Sintesi delle prove pubbliche

Adattamento del modello locale per Mac mini e Mac Studio

Uno strumento di pianificazione per gli ultimi rappresentanti a peso aperto Qwen3.8, DeepSeek V4, GLM-5.3 e Gemma 4. Inizia con i dati dei file pubblici e i calcoli della capacità, con tempi di esecuzione separati e prove del lavoro svolto, quindi decidi se è giustificata una sostituzione.

Ultima verifica: 28 agosto 2026 · Supporto decisionale indipendente, non consiglio Apple

Risposta breve

Questa pagina mette a confronto i rappresentanti di ciascuna famiglia attualmente di interesse. Le versioni precedenti vengono conservate solo come record di compatibilità ed escluse dalla matrice principale. Anche se il modello rientra nella memoria, il richiamo dello strumento, JSON, il comportamento visivo, la concorrenza e il comportamento persistente potrebbero essere sconosciuti.

strumenti di dialogo

Controlla se il modello è compatibile con il tuo Mac.

Scegli il tuo modello o la configurazione del Mac per visualizzare budget ponderati, spazio di memoria, prove di runtime e passaggi successivi al nostro Advisor gratuito.

Apri l'esploratore di idoneità

Sintesi delle prove pubbliche, non un benchmark di prima parte.

Questo strumento di pianificazione combina schede modello ufficiali, metadati di file GGUF/QAT pubblici, documentazione di runtime, valutazioni pubblicate e report della comunità chiaramente etichettati. Non afferma che Keep o Upgrade eseguissero tutti i modelli e Apple o i fornitori di modelli non approvano questa pagina.

Scarica CSV Scarica JSON · Stessi dati di origine di questa pagina

I download sono istantanee facilmente citabili con URL di origine, livelli di evidenza e data dei dati. Link a questo Hub come spiegazione canonica; gli endpoint di download rimangono fuori dalla mappa del sito.

Confronta con una configurazione Mac

Mac mini M6 · 16GB153GB/s · US$899 · prezzo base

Matrice delle prove del modello

Modelli a peso aperto più recenti e di grande interesse

La matrice primaria dà la priorità alle attuali versioni di punta e agli utili rappresentanti della pianificazione Mac. La capacità utilizza la dimensione dell'artefatto pubblico selezionato quando disponibile; i numeri basati su parametri sono solo stime di pianificazione. Espandi Fonti e limiti su qualsiasi riga per controllare i segnali della community attribuibili e i relativi limiti.

Famiglia/livello	Modello e versione	Quantizzazione/artefatto	Capacità sul Mac selezionato	Prove di runtime	Affidabilità	Dettagli
QwenPiccolo	Qwen3.8 27BQwen3.8 · 27B dense	Q4_0 · GGUF16.1 GB · Public file metadata	Non entra-8.06GB rimanenti dopo la riserva	Supporto ufficialemlx2 segnali social · vedi fonti	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public file metadata16.1 GB · exact public file metadata Fonte del runtimemlx · Supporto ufficiale Prove della comunità (2)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto. Reddit · Qwen3.8-27B on a 24GB M4 Pro Mac miniMac mini · M4 Pro · 24GB · llama.cpp b10488 · GGUF Q4_K_M / IQ4_XSQ4_K_M 17.77GB; about 11.4 tok/s decode; about 16.6GB resident with q8 KV at 32K context · medium confidenceA first-person report says the 27B model is usable on 24GB, but the memory ceiling is visible once context and KV cache are included. Limite: One user, one build, and one workload; not a Keep or Upgrade test. Tech media / blog · Qwen3.8-27B 4-bit on a 32GB M1 ProMacBook Pro · M1 Pro · 32GB · MLX-VLM · MLX 4-bit8–8.7 tok/s; weights about 16.05GB; peak memory about 18.5–21.7GB · medium confidenceA detailed local deployment log reports a stable 32GB starting point and explicitly discourages 16GB for this 27B representative. Limite: Community log with different thermals, context, and prompt mix from the product matrix. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
QwenPrincipale	Qwen3.8-Flash-Next 180B (6B active)Qwen3.8-Flash-Next · 180B total / 6B active	BF16 · Safetensors360.0 GB · Public file metadata	Non entra-352GB rimanenti dopo la riserva	Report della comunitàmlx1 segnale social · vedi fonti	mediumVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public file metadata360.0 GB · exact public file metadata Fonte del runtimemlx · Report della comunità Prove della comunità (1)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto. Reddit · Latest-model comparison thread (non-Apple hardware)4× DGX Spark / other non-Apple systemsUsers discuss Qwen3.8 Flash as an exploration/subagent model; no Mac tuple or reproducible throughput is provided. · low confidenceThis is a freshness and interest signal only, not evidence that Flash Next fits a Mac configuration. Limite: Cross-hardware discussion; excluded from Mac capacity conclusions. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
QwenConfine	Qwen3.8 2.4T-A95BQwen3.8 · 2400B total / 95B active	Q4_K_M · GGUF1440.0 GB · Public model source; exact file not registered	Non entra-1432GB rimanenti dopo la riserva	Supporto ufficialemlx	mediumVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public model source; exact file not registered1440.0 GB · parameter planning estimate Fonte del runtimemlx · Supporto ufficiale In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
DeepSeekPrincipale	DeepSeek-V4-Flash 284B-A13BDeepSeek V4 · 284B total / 13B active	MLX-4bit · MLX170.4 GB · Public model source; exact file not registered	Non entra-162.4GB rimanenti dopo la riserva	Report della comunitàmlx7 segnali social · vedi fonti	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public model source; exact file not registered170.4 GB · parameter planning estimate Fonte del runtimemlx · Report della comunità Prove della comunità (7)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto. Reddit · DeepSeek V4 Flash on an M5 Max 128GBMac · Apple M5 Max · 128GB · ds4 / SSD-streaming path · DwarfStar IQ2XXSAbout 81GB model working set and about 31.06 tok/s in the reported run · medium confidenceA direct Apple Silicon report shows a practical 128GB path for the Flash representative. Limite: Community benchmark; exact context, thermal state, and build can change the result. Tech media / blog · DeepSeek V4 Flash on an M3 Max 128GBMacBook Pro · M3 Max · 128GB · ds4 experimental forkAbout 21 tok/s and about 81GB resident in the reported run · medium confidenceAn independent technical write-up corroborates that 128GB-class Apple Silicon can run a local Flash path. Limite: The article is tied to an experimental fork and does not validate mainline Ollama or llama.cpp. Reddit · DeepSeek V4 Flash on 64GB Macs with SSD streaming64GB Apple Silicon Macs · 64GB · ds4 SSD streaming · IQ2XXS / 2-bit classReports cluster around roughly 10–15 tok/s; technically possible, but not a comfortable in-memory workflow · medium confidence64GB appears technically viable only with aggressive quantization and storage streaming, so it is kept as a constrained edge case. Limite: Mixed community reports and different Mac generations; do not read this as a 64GB recommendation. GitHub · MLX DeepSeek V4 residency growth and resource-limit crashM4 Max 128GB and other Apple Silicon reports · 128GB · MLX-LMReproduction reports deterministic resource-limit failure around 11,300 generated tokens · high confidenceThe issue is important negative evidence: a model can fit and still fail on long generations because of runtime memory behavior. Limite: Issue state and fixes can change; check the runtime version and referenced patch before relying on MLX. GitHub · MLX cache/residency failure on an M3 Ultra 512GBMac Studio · M3 Ultra 512GB · 512GB · MLX-LMReports failure around 11,456–11,488 generated tokens in a production-style run · high confidenceEven very large unified-memory systems are not automatically stable for long-context MLX runs. Limite: This is a runtime bug report, not a capacity limit; a patched release may change the result. GitHub · DeepSeek V4 Flash Q8 garbled output on Apple NEONApple Silicon CPU / NEON · llama.cpp · Q8A specific repack path produced garbled output · high confidenceThis narrows the risk to a quantization/repack/runtime combination instead of treating every llama.cpp path as equivalent. Limite: Do not generalize one broken Q8 repack to all DeepSeek V4 Flash formats. GitHub · ds4 Apple Silicon throughput tableMac Studio · M3 Ultra 512GB · 512GB · ds4 Metal/CUDA engine · Q4 classRepository table reports short/long generation figures around 78.95 / 35.50 tok/s · medium confidenceA purpose-built engine provides a useful upper-bound signal for a large-memory Mac Studio, but it is not a mainstream runtime guarantee. Limite: Single graph worker, no batching, and engine-specific optimizations; compare only directionally. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
DeepSeekConfine	DeepSeek-V4-Pro 1.6T-A49BDeepSeek V4 · 1600B total / 49B active	MLX-4bit · MLX960.0 GB · Public model source; exact file not registered	Non entra-952GB rimanenti dopo la riserva	Sconosciutollama.cpp	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public model source; exact file not registered960.0 GB · parameter planning estimate Fonte del runtimellama.cpp · Sconosciuto In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
GLMPrincipale	GLM-5.3-Flash 320B-A18BGLM-5.3-Flash · 320B total / 18B active	MLX-4bit · MLX204.0 GB · Public file metadata	Non entra-195.99GB rimanenti dopo la riserva	Report della comunitàmlx2 segnali social · vedi fonti	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public file metadata204.0 GB · exact public file metadata Fonte del runtimemlx · Report della comunità Prove della comunità (2)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto. Hugging Face · Community GLM-5.3-Flash MLX quantization ladderMLX · 2bit-lite / 2 / 3 / 4 / 6-bit MLXPublic metadata lists about 102.43GB to 295.60GB per variant; the 4-bit root mirror is about 203.99GB · medium confidenceThe newer repository makes the precision trade-off explicit and includes a 2bit-lite path, but it does not prove loadability or speed on a particular Mac. Limite: Community conversion; full-repository size includes duplicate root and variant folders, and no trustworthy Apple first-person benchmark is attached. Reddit · GLM-5.3 Flash comparison thread (non-Apple hardware)Dual DGX Spark / other non-Apple systemsOne report describes roughly 22 tok/s on dual Spark and a slower, more verbose interaction style · low confidenceThis indicates active community comparison but cannot be translated into a Mac recommendation. Limite: Cross-hardware, anecdotal, and workload-specific. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
GLMConfine	GLM-5.3 744B-A40BGLM-5.3 · 744B total / 40B active	MLX-4bit · MLX446.4 GB · Public model source; exact file not registered	Non entra-438.4GB rimanenti dopo la riserva	Sconosciutollama.cpp	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public model source; exact file not registered446.4 GB · parameter planning estimate Fonte del runtimellama.cpp · Sconosciuto In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
Google GemmaPiccolo	Gemma 4 E2BGemma 4 · 2.3B effective / 5.1B total · PLE	Q4_0 · QAT3.35 GB · Public file metadata	Con margine4.65GB rimanenti dopo la riserva	Report della comunitàllama.cpp1 segnale social · vedi fonti	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public file metadata3.35 GB · exact public file metadata Fonte del runtimellama.cpp · Report della comunità Prove della comunità (1)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto. Tech media / blog · Apple Silicon local-AI comparison including Gemma 4Mac Studio / M4 Max 128GB test systems · 128GB · llama.cppGemma 4 appears in the comparison; the article warns that bandwidth is not the sole predictor and does not publish this exact matrix tuple · low confidenceUseful context for Apple Silicon behavior, but not an exact Gemma 4 E2B fit or performance claim. Limite: Different model/quantization details and media test methodology from this product matrix. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
Google GemmaPrincipale	Gemma 4 12B UnifiedGemma 4 · 11.95B dense	Q4_0 · QAT6.98 GB · Public file metadata	Stretto1.02GB rimanenti dopo la riserva	Sconosciutollama.cpp	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public file metadata6.98 GB · exact public file metadata Fonte del runtimellama.cpp · Sconosciuto In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito
Google GemmaConfine	Gemma 4 26B-A4BGemma 4 · 25.2B total / 3.8B active	Q4_0 · QAT14.4 GB · Public file metadata	Non entra-6.44GB rimanenti dopo la riserva	Report della comunitàllama.cpp	highVerificato 2026-08-28	Fonti e limiti Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28 Public file metadata14.4 GB · exact public file metadata Fonte del runtimellama.cpp · Report della comunità In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime. Public file metadata or a parameter estimate is not a first-party benchmark. Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing. Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result. Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks. The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here. Utilizzare nel consulente gratuito

Qwen · Piccolo

Qwen3.8 27B

Qwen3.8 · 27B dense

Non entra

Quantizzazione/artefatto: Q4_0 · GGUF16.1 GB · Public file metadata
Capacità sul Mac selezionato: -8.06GB rimanenti dopo la riserva
Prove di runtime: Supporto ufficialemlx
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public file metadata16.1 GB · exact public file metadata
Fonte del runtimemlx · Supporto ufficiale
Prove della comunità (2)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto.
Reddit · Qwen3.8-27B on a 24GB M4 Pro Mac miniMac mini · M4 Pro · 24GB · llama.cpp b10488 · GGUF Q4_K_M / IQ4_XSQ4_K_M 17.77GB; about 11.4 tok/s decode; about 16.6GB resident with q8 KV at 32K context · medium confidenceA first-person report says the 27B model is usable on 24GB, but the memory ceiling is visible once context and KV cache are included. Limite: One user, one build, and one workload; not a Keep or Upgrade test.
Tech media / blog · Qwen3.8-27B 4-bit on a 32GB M1 ProMacBook Pro · M1 Pro · 32GB · MLX-VLM · MLX 4-bit8–8.7 tok/s; weights about 16.05GB; peak memory about 18.5–21.7GB · medium confidenceA detailed local deployment log reports a stable 32GB starting point and explicitly discourages 16GB for this 27B representative. Limite: Community log with different thermals, context, and prompt mix from the product matrix.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

Qwen · Principale

Qwen3.8-Flash-Next 180B (6B active)

Qwen3.8-Flash-Next · 180B total / 6B active

Non entra

Quantizzazione/artefatto: BF16 · Safetensors360.0 GB · Public file metadata
Capacità sul Mac selezionato: -352GB rimanenti dopo la riserva
Prove di runtime: Report della comunitàmlx
Affidabilità: mediumVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public file metadata360.0 GB · exact public file metadata
Fonte del runtimemlx · Report della comunità
Prove della comunità (1)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto.
Reddit · Latest-model comparison thread (non-Apple hardware)4× DGX Spark / other non-Apple systemsUsers discuss Qwen3.8 Flash as an exploration/subagent model; no Mac tuple or reproducible throughput is provided. · low confidenceThis is a freshness and interest signal only, not evidence that Flash Next fits a Mac configuration. Limite: Cross-hardware discussion; excluded from Mac capacity conclusions.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

Qwen · Confine

Qwen3.8 2.4T-A95B

Qwen3.8 · 2400B total / 95B active

Non entra

Quantizzazione/artefatto: Q4_K_M · GGUF1440.0 GB · Public model source; exact file not registered
Capacità sul Mac selezionato: -1432GB rimanenti dopo la riserva
Prove di runtime: Supporto ufficialemlx
Affidabilità: mediumVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public model source; exact file not registered1440.0 GB · parameter planning estimate
Fonte del runtimemlx · Supporto ufficiale
In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

DeepSeek · Principale

DeepSeek-V4-Flash 284B-A13B

DeepSeek V4 · 284B total / 13B active

Non entra

Quantizzazione/artefatto: MLX-4bit · MLX170.4 GB · Public model source; exact file not registered
Capacità sul Mac selezionato: -162.4GB rimanenti dopo la riserva
Prove di runtime: Report della comunitàmlx
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public model source; exact file not registered170.4 GB · parameter planning estimate
Fonte del runtimemlx · Report della comunità
Prove della comunità (7)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto.
Reddit · DeepSeek V4 Flash on an M5 Max 128GBMac · Apple M5 Max · 128GB · ds4 / SSD-streaming path · DwarfStar IQ2XXSAbout 81GB model working set and about 31.06 tok/s in the reported run · medium confidenceA direct Apple Silicon report shows a practical 128GB path for the Flash representative. Limite: Community benchmark; exact context, thermal state, and build can change the result.
Tech media / blog · DeepSeek V4 Flash on an M3 Max 128GBMacBook Pro · M3 Max · 128GB · ds4 experimental forkAbout 21 tok/s and about 81GB resident in the reported run · medium confidenceAn independent technical write-up corroborates that 128GB-class Apple Silicon can run a local Flash path. Limite: The article is tied to an experimental fork and does not validate mainline Ollama or llama.cpp.
Reddit · DeepSeek V4 Flash on 64GB Macs with SSD streaming64GB Apple Silicon Macs · 64GB · ds4 SSD streaming · IQ2XXS / 2-bit classReports cluster around roughly 10–15 tok/s; technically possible, but not a comfortable in-memory workflow · medium confidence64GB appears technically viable only with aggressive quantization and storage streaming, so it is kept as a constrained edge case. Limite: Mixed community reports and different Mac generations; do not read this as a 64GB recommendation.
GitHub · MLX DeepSeek V4 residency growth and resource-limit crashM4 Max 128GB and other Apple Silicon reports · 128GB · MLX-LMReproduction reports deterministic resource-limit failure around 11,300 generated tokens · high confidenceThe issue is important negative evidence: a model can fit and still fail on long generations because of runtime memory behavior. Limite: Issue state and fixes can change; check the runtime version and referenced patch before relying on MLX.
GitHub · MLX cache/residency failure on an M3 Ultra 512GBMac Studio · M3 Ultra 512GB · 512GB · MLX-LMReports failure around 11,456–11,488 generated tokens in a production-style run · high confidenceEven very large unified-memory systems are not automatically stable for long-context MLX runs. Limite: This is a runtime bug report, not a capacity limit; a patched release may change the result.
GitHub · DeepSeek V4 Flash Q8 garbled output on Apple NEONApple Silicon CPU / NEON · llama.cpp · Q8A specific repack path produced garbled output · high confidenceThis narrows the risk to a quantization/repack/runtime combination instead of treating every llama.cpp path as equivalent. Limite: Do not generalize one broken Q8 repack to all DeepSeek V4 Flash formats.
GitHub · ds4 Apple Silicon throughput tableMac Studio · M3 Ultra 512GB · 512GB · ds4 Metal/CUDA engine · Q4 classRepository table reports short/long generation figures around 78.95 / 35.50 tok/s · medium confidenceA purpose-built engine provides a useful upper-bound signal for a large-memory Mac Studio, but it is not a mainstream runtime guarantee. Limite: Single graph worker, no batching, and engine-specific optimizations; compare only directionally.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

DeepSeek · Confine

DeepSeek-V4-Pro 1.6T-A49B

DeepSeek V4 · 1600B total / 49B active

Non entra

Quantizzazione/artefatto: MLX-4bit · MLX960.0 GB · Public model source; exact file not registered
Capacità sul Mac selezionato: -952GB rimanenti dopo la riserva
Prove di runtime: Sconosciutollama.cpp
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public model source; exact file not registered960.0 GB · parameter planning estimate
Fonte del runtimellama.cpp · Sconosciuto
In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

GLM · Principale

GLM-5.3-Flash 320B-A18B

GLM-5.3-Flash · 320B total / 18B active

Non entra

Quantizzazione/artefatto: MLX-4bit · MLX204.0 GB · Public file metadata
Capacità sul Mac selezionato: -195.99GB rimanenti dopo la riserva
Prove di runtime: Report della comunitàmlx
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public file metadata204.0 GB · exact public file metadata
Fonte del runtimemlx · Report della comunità
Prove della comunità (2)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto.
Hugging Face · Community GLM-5.3-Flash MLX quantization ladderMLX · 2bit-lite / 2 / 3 / 4 / 6-bit MLXPublic metadata lists about 102.43GB to 295.60GB per variant; the 4-bit root mirror is about 203.99GB · medium confidenceThe newer repository makes the precision trade-off explicit and includes a 2bit-lite path, but it does not prove loadability or speed on a particular Mac. Limite: Community conversion; full-repository size includes duplicate root and variant folders, and no trustworthy Apple first-person benchmark is attached.
Reddit · GLM-5.3 Flash comparison thread (non-Apple hardware)Dual DGX Spark / other non-Apple systemsOne report describes roughly 22 tok/s on dual Spark and a slower, more verbose interaction style · low confidenceThis indicates active community comparison but cannot be translated into a Mac recommendation. Limite: Cross-hardware, anecdotal, and workload-specific.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

GLM · Confine

GLM-5.3 744B-A40B

GLM-5.3 · 744B total / 40B active

Non entra

Quantizzazione/artefatto: MLX-4bit · MLX446.4 GB · Public model source; exact file not registered
Capacità sul Mac selezionato: -438.4GB rimanenti dopo la riserva
Prove di runtime: Sconosciutollama.cpp
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public model source; exact file not registered446.4 GB · parameter planning estimate
Fonte del runtimellama.cpp · Sconosciuto
In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

Google Gemma · Piccolo

Gemma 4 E2B

Gemma 4 · 2.3B effective / 5.1B total · PLE

Con margine

Quantizzazione/artefatto: Q4_0 · QAT3.35 GB · Public file metadata
Capacità sul Mac selezionato: 4.65GB rimanenti dopo la riserva
Prove di runtime: Report della comunitàllama.cpp
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public file metadata3.35 GB · exact public file metadata
Fonte del runtimellama.cpp · Report della comunità
Prove della comunità (1)I report pubblici sono solo segnali e non trasformano mai questa pagina in un test del progetto.
Tech media / blog · Apple Silicon local-AI comparison including Gemma 4Mac Studio / M4 Max 128GB test systems · 128GB · llama.cppGemma 4 appears in the comparison; the article warns that bandwidth is not the sole predictor and does not publish this exact matrix tuple · low confidenceUseful context for Apple Silicon behavior, but not an exact Gemma 4 E2B fit or performance claim. Limite: Different model/quantization details and media test methodology from this product matrix.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

Google Gemma · Principale

Gemma 4 12B Unified

Gemma 4 · 11.95B dense

Stretto

Quantizzazione/artefatto: Q4_0 · QAT6.98 GB · Public file metadata
Capacità sul Mac selezionato: 1.02GB rimanenti dopo la riserva
Prove di runtime: Sconosciutollama.cpp
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public file metadata6.98 GB · exact public file metadata
Fonte del runtimellama.cpp · Sconosciuto
In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

Google Gemma · Confine

Gemma 4 26B-A4B

Gemma 4 · 25.2B total / 3.8B active

Non entra

Quantizzazione/artefatto: Q4_0 · QAT14.4 GB · Public file metadata
Capacità sul Mac selezionato: -6.44GB rimanenti dopo la riserva
Prove di runtime: Report della comunitàllama.cpp
Affidabilità: highVerificato 2026-08-28

Fonti e limiti

Fonte ufficiale del modelloModello family, parameters, and release context · Verificato 2026-08-28
Public file metadata14.4 GB · exact public file metadata
Fonte del runtimellama.cpp · Report della comunità
In questo aggiornamento non è stato acquisito alcun report Mac di prima mano affidabile.La capacità resta una stima di pianificazione basata su metadati pubblici; l'assenza di prove non convalida il runtime.
Public file metadata or a parameter estimate is not a first-party benchmark.
Community artifacts and reports do not imply vendor endorsement or Keep or Upgrade testing.
Social evidence is directional: build, context, thermals, storage streaming, and prompt mix can change the result.
Context, concurrency, tool calling, JSON, vision, and sustained behavior remain separate capability checks.
The primary matrix prioritizes current flagship releases; legacy versions remain compatibility records but are intentionally omitted here.

Utilizzare nel consulente gratuito

Prova di capacità

La capacità del flusso di lavoro è una questione separata

Le finestre di contesto sono metadati del modello. La concorrenza, le chiamate agli strumenti, l'affidabilità JSON, la visione e il comportamento sostenuto dipendono dal modello esatto, dal runtime, dal client, dal carico di memoria e dall'attività. Una cella positiva non è ancora una garanzia per il tuo flusso di lavoro.

Modello	Contesto	Concorrenza	Chiamate agli strumenti	JSON	Visione	Uso prolungato
Qwen3.8 27BPiccolo	Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Supporto ufficialeThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	SconosciutoVision support is not assumed from family branding or an unrelated model variant.	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Qwen3.8-Flash-Next 180B (6B active)Principale	Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Supporto ufficialeThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	Supporto ufficialeThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Qwen3.8 2.4T-A95BConfine	Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Supporto ufficialeThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	SconosciutoVision support is not assumed from family branding or an unrelated model variant.	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
DeepSeek-V4-Flash 284B-A13BPrincipale	Supporto ufficialePublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Supporto ufficialeThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	SconosciutoVision support is not assumed from family branding or an unrelated model variant.	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
DeepSeek-V4-Pro 1.6T-A49BConfine	Supporto ufficialePublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Supporto ufficialeThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	SconosciutoVision support is not assumed from family branding or an unrelated model variant.	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
GLM-5.3-Flash 320B-A18BPrincipale	Supporto ufficialePublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Supporto ufficialeThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte	Benchmark pubblicoPublic GLM evaluations cover agentic or structured tasks, but they do not establish JSON validity in every client.Fonte	Supporto ufficialeThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
GLM-5.3 744B-A40BConfine	Supporto ufficialePublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Supporto ufficialeThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	SconosciutoVision support is not assumed from family branding or an unrelated model variant.	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Gemma 4 E2BPiccolo	Supporto ufficialePublic model metadata lists 131,072 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	SconosciutoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	Supporto ufficialeThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Gemma 4 12B UnifiedPrincipale	Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	SconosciutoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	SconosciutoVision support is not assumed from family branding or an unrelated model variant.	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Gemma 4 26B-A4BConfine	Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	SconosciutoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.	SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	SconosciutoVision support is not assumed from family branding or an unrelated model variant.	SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Qwen3.8 27BPiccolo

Contesto: Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: Supporto ufficialeThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: SconosciutoVision support is not assumed from family branding or an unrelated model variant.
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Qwen3.8-Flash-Next 180B (6B active)Principale

Contesto: Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: Supporto ufficialeThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: Supporto ufficialeThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Qwen3.8 2.4T-A95BConfine

Contesto: Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: Supporto ufficialeThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: SconosciutoVision support is not assumed from family branding or an unrelated model variant.
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

DeepSeek-V4-Flash 284B-A13BPrincipale

Contesto: Supporto ufficialePublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: Supporto ufficialeThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: SconosciutoVision support is not assumed from family branding or an unrelated model variant.
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

DeepSeek-V4-Pro 1.6T-A49BConfine

Contesto: Supporto ufficialePublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: Supporto ufficialeThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: SconosciutoVision support is not assumed from family branding or an unrelated model variant.
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

GLM-5.3-Flash 320B-A18BPrincipale

Contesto: Supporto ufficialePublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: Supporto ufficialeThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte
JSON: Benchmark pubblicoPublic GLM evaluations cover agentic or structured tasks, but they do not establish JSON validity in every client.Fonte
Visione: Supporto ufficialeThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

GLM-5.3 744B-A40BConfine

Contesto: Supporto ufficialePublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: Supporto ufficialeThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: SconosciutoVision support is not assumed from family branding or an unrelated model variant.
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Gemma 4 E2BPiccolo

Contesto: Supporto ufficialePublic model metadata lists 131,072 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: SconosciutoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: Supporto ufficialeThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Gemma 4 12B UnifiedPrincipale

Contesto: Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: SconosciutoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: SconosciutoVision support is not assumed from family branding or an unrelated model variant.
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Gemma 4 26B-A4BConfine

Contesto: Supporto ufficialePublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrenza: Inferenza di sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chiamate agli strumenti: SconosciutoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.
JSON: SconosciutoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visione: SconosciutoVision support is not assumed from family branding or an unrelated model variant.
Uso prolungato: SconosciutoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Come leggere le etichette

Supporto ufficiale La documentazione ufficiale del modello/runtime descrive il percorso della capacità; il comportamento esatto del client va ancora verificato.
Benchmark pubblico Esiste una valutazione pubblicata, ma non è un benchmark di throughput o workflow su Mac.
Report della comunità I report pubblici o gli artefatti della comunità sono segnali, non test del progetto né garanzie del fornitore.
Inferenza di sistema Inferenza delimitata dall'architettura di capacità/runtime, non un risultato osservato.
Sconosciuto Non è stata acquisita una prova pubblica affidabile per questa affermazione esatta.
Verificato dal progetto È verificata solo la combinazione esatta di modello, quantizzazione, runtime e macchina nel registro smoke del progetto.

Porta il tuo vero carico di lavoro

Trasformare le prove pubbliche in una decisione personale.

L'Advisor gratuito aggiunge i tuoi attuali requisiti Mac, contesto, concorrenza, budget, compatibilità e offline. Si può dire “è necessario verificare” quando le prove pubbliche non sono sufficienti.

Esegui il consulente gratuito

Domande frequenti

Controllare i limiti delle prove prima dell'acquisto.

Si tratta di un benchmark Keep o Upgrade?

No. Uno strumento di sintesi e pianificazione delle prove pubbliche. Solo la combinazione esatta verificata nei test di base (smoke test) del progetto verrà mostrata come verificata.

Come calcoli la capacità del tuo Mac?

Utilizza la dimensione del file pubblico selezionata se è disponibile il file esatto o il totale dei file suddivisi, altrimenti utilizza la stima di pianificazione esplicita. Il budget del modello riserva spazio a macOS, runtime e contesto.

MoE richiede meno memoria se è un parametro valido?

No. I pesi totali memorizzati rappresentano la condizione di capacità. I parametri validi descrivono solo la quantità di calcolo per token.

La lunghezza del contesto ufficiale funziona su qualsiasi Mac?

No. La lunghezza del contesto corrisponde ai metadati del modello. Oltre a peso, runtime, sistema e altre app, devi anche avere memoria sufficiente sul Mac selezionato.

Si può comprare un Mac semplicemente guardando la tabella?

Dopo aver ristretto l'ambito della tua pianificazione con questa matrice, inserisci il tuo Mac attuale, il contesto, la concorrenza, i clienti, il budget, la compatibilità e i requisiti offline nel nostro Advisor gratuito. La prova pubblica non è una garanzia di acquisto.

portare il lavoro vero e proprio

Utilizza prove pubbliche per informare le tue decisioni di acquisto.

Free Advisor aggiunge il tuo attuale Mac, contesto, concorrenza, budget, compatibilità, requisiti offline e indica che è necessaria la verifica se non ci sono prove pubbliche sufficienti.

Avvia consulenza gratuita

Keep or Upgrade è uno strumento indipendente; Apple non sponsorizza né approva questa pagina.