Síntese de evidências públicas

Adaptação de modelo local para Mac mini e Mac Studio

Uma ferramenta de planejamento para os mais recentes representantes de peso aberto Qwen3.8, DeepSeek V4, GLM-5.3 e Gemma 4. Comece com fatos de arquivos públicos e cálculos de capacidade, separe o tempo de execução e as evidências de trabalho e, em seguida, decida se uma substituição é necessária.

Última revisão: 28 de agosto de 2026 · Apoio independente, não conselho da Apple

Conclusão breve

Esta página compara os representantes de cada família que atualmente interessam. As versões anteriores são mantidas apenas como registro de compatibilidade e excluídas da matriz principal. Mesmo que o modelo caiba na memória, a invocação da ferramenta, o JSON, o visual, a simultaneidade e o comportamento persistente podem ser desconhecidos.

ferramentas de diálogo

Verifique se o modelo é compatível com o seu Mac.

Escolha seu modelo ou configuração de Mac para ver orçamentos ponderados, espaço de memória, evidências de tempo de execução e próximas etapas do nosso Consultor gratuito.

Abrir o explorador de adequação

Síntese de evidências públicas, não um benchmark primário.

Esta ferramenta de planejamento combina cartões modelo oficiais, metadados de arquivos públicos GGUF/QAT, documentação de tempo de execução, avaliações publicadas e relatórios comunitários claramente identificados. Não afirma que Keep ou Upgrade executou todos os modelos, e a Apple ou os fornecedores de modelos não endossam esta página.

Baixar CSV Descarregar JSON · Os mesmos dados de origem desta página

Os downloads são instantâneos fáceis de citar com URLs de origem, níveis de evidência e data dos dados. Link para este Hub como explicação canônica; os endpoints de download ficam fora do mapa do site.

Compare com uma configuração Mac

Mac mini M6 · 16GB153GB/s · US$899 · preço inicial

Matriz de evidências do modelo

Modelos abertos mais recentes e de alto interesse

A matriz primária prioriza os principais lançamentos atuais e representantes úteis de planejamento do Mac. A capacidade usa o tamanho do artefato público selecionado quando disponível; os números baseados em parâmetros são apenas estimativas de planejamento. Expanda Fontes e limites em qualquer linha para inspecionar sinais de comunidade atribuíveis e seus limites.

Família / nível	Modelo e versão	Quantização/artefato	Capacidade no Mac selecionado	Evidência de tempo de execução	Confiança	Detalhes
QwenPequeno	Qwen3.8 27BQwen3.8 · 27B dense	Q4_0 · GGUF16.1 GB · Public file metadata	Não cabe-8.06GB restantes após a reserva	Suporte oficialmlx2 sinais sociais · ver fontes	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public file metadata16.1 GB · exact public file metadata Fonte do runtimemlx · Suporte oficial Evidência da comunidade (2)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto. 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. Limitação: 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. Limitação: 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. Use no consultor gratuito
QwenPrincipal	Qwen3.8-Flash-Next 180B (6B active)Qwen3.8-Flash-Next · 180B total / 6B active	BF16 · Safetensors360.0 GB · Public file metadata	Não cabe-352GB restantes após a reserva	Relato da comunidademlx1 sinal social · ver fontes	mediumVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public file metadata360.0 GB · exact public file metadata Fonte do runtimemlx · Relato da comunidade Evidência da comunidade (1)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto. 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. Limitação: 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. Use no consultor gratuito
QwenLimite	Qwen3.8 2.4T-A95BQwen3.8 · 2400B total / 95B active	Q4_K_M · GGUF1440.0 GB · Public model source; exact file not registered	Não cabe-1432GB restantes após a reserva	Suporte oficialmlx	mediumVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public model source; exact file not registered1440.0 GB · parameter planning estimate Fonte do runtimemlx · Suporte oficial Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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. Use no consultor gratuito
DeepSeekPrincipal	DeepSeek-V4-Flash 284B-A13BDeepSeek V4 · 284B total / 13B active	MLX-4bit · MLX170.4 GB · Public model source; exact file not registered	Não cabe-162.4GB restantes após a reserva	Relato da comunidademlx7 sinais sociais · ver fontes	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public model source; exact file not registered170.4 GB · parameter planning estimate Fonte do runtimemlx · Relato da comunidade Evidência da comunidade (7)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto. 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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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. Use no consultor gratuito
DeepSeekLimite	DeepSeek-V4-Pro 1.6T-A49BDeepSeek V4 · 1600B total / 49B active	MLX-4bit · MLX960.0 GB · Public model source; exact file not registered	Não cabe-952GB restantes após a reserva	Desconhecidollama.cpp	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public model source; exact file not registered960.0 GB · parameter planning estimate Fonte do runtimellama.cpp · Desconhecido Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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. Use no consultor gratuito
GLMPrincipal	GLM-5.3-Flash 320B-A18BGLM-5.3-Flash · 320B total / 18B active	MLX-4bit · MLX204.0 GB · Public file metadata	Não cabe-195.99GB restantes após a reserva	Relato da comunidademlx2 sinais sociais · ver fontes	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public file metadata204.0 GB · exact public file metadata Fonte do runtimemlx · Relato da comunidade Evidência da comunidade (2)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto. 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. Limitação: 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. Limitação: 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. Use no consultor gratuito
GLMLimite	GLM-5.3 744B-A40BGLM-5.3 · 744B total / 40B active	MLX-4bit · MLX446.4 GB · Public model source; exact file not registered	Não cabe-438.4GB restantes após a reserva	Desconhecidollama.cpp	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public model source; exact file not registered446.4 GB · parameter planning estimate Fonte do runtimellama.cpp · Desconhecido Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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. Use no consultor gratuito
Google GemmaPequeno	Gemma 4 E2BGemma 4 · 2.3B effective / 5.1B total · PLE	Q4_0 · QAT3.35 GB · Public file metadata	Com margem4.65GB restantes após a reserva	Relato da comunidadellama.cpp1 sinal social · ver fontes	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public file metadata3.35 GB · exact public file metadata Fonte do runtimellama.cpp · Relato da comunidade Evidência da comunidade (1)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto. 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. Limitação: 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. Use no consultor gratuito
Google GemmaPrincipal	Gemma 4 12B UnifiedGemma 4 · 11.95B dense	Q4_0 · QAT6.98 GB · Public file metadata	Ajustado1.02GB restantes após a reserva	Desconhecidollama.cpp	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public file metadata6.98 GB · exact public file metadata Fonte do runtimellama.cpp · Desconhecido Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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. Use no consultor gratuito
Google GemmaLimite	Gemma 4 26B-A4BGemma 4 · 25.2B total / 3.8B active	Q4_0 · QAT14.4 GB · Public file metadata	Não cabe-6.44GB restantes após a reserva	Relato da comunidadellama.cpp	highVerificado 2026-08-28	Fontes e limites Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28 Public file metadata14.4 GB · exact public file metadata Fonte do runtimellama.cpp · Relato da comunidade Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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. Use no consultor gratuito

Qwen · Pequeno

Qwen3.8 27B

Qwen3.8 · 27B dense

Não cabe

Quantização/artefato: Q4_0 · GGUF16.1 GB · Public file metadata
Capacidade no Mac selecionado: -8.06GB restantes após a reserva
Evidência de tempo de execução: Suporte oficialmlx
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public file metadata16.1 GB · exact public file metadata
Fonte do runtimemlx · Suporte oficial
Evidência da comunidade (2)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto.
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. Limitação: 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. Limitação: 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.

Use no consultor gratuito

Qwen · Principal

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

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

Não cabe

Quantização/artefato: BF16 · Safetensors360.0 GB · Public file metadata
Capacidade no Mac selecionado: -352GB restantes após a reserva
Evidência de tempo de execução: Relato da comunidademlx
Confiança: mediumVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public file metadata360.0 GB · exact public file metadata
Fonte do runtimemlx · Relato da comunidade
Evidência da comunidade (1)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto.
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. Limitação: 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.

Use no consultor gratuito

Qwen · Limite

Qwen3.8 2.4T-A95B

Qwen3.8 · 2400B total / 95B active

Não cabe

Quantização/artefato: Q4_K_M · GGUF1440.0 GB · Public model source; exact file not registered
Capacidade no Mac selecionado: -1432GB restantes após a reserva
Evidência de tempo de execução: Suporte oficialmlx
Confiança: mediumVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public model source; exact file not registered1440.0 GB · parameter planning estimate
Fonte do runtimemlx · Suporte oficial
Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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.

Use no consultor gratuito

DeepSeek · Principal

DeepSeek-V4-Flash 284B-A13B

DeepSeek V4 · 284B total / 13B active

Não cabe

Quantização/artefato: MLX-4bit · MLX170.4 GB · Public model source; exact file not registered
Capacidade no Mac selecionado: -162.4GB restantes após a reserva
Evidência de tempo de execução: Relato da comunidademlx
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public model source; exact file not registered170.4 GB · parameter planning estimate
Fonte do runtimemlx · Relato da comunidade
Evidência da comunidade (7)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto.
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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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. Limitação: 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.

Use no consultor gratuito

DeepSeek · Limite

DeepSeek-V4-Pro 1.6T-A49B

DeepSeek V4 · 1600B total / 49B active

Não cabe

Quantização/artefato: MLX-4bit · MLX960.0 GB · Public model source; exact file not registered
Capacidade no Mac selecionado: -952GB restantes após a reserva
Evidência de tempo de execução: Desconhecidollama.cpp
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public model source; exact file not registered960.0 GB · parameter planning estimate
Fonte do runtimellama.cpp · Desconhecido
Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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.

Use no consultor gratuito

GLM · Principal

GLM-5.3-Flash 320B-A18B

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

Não cabe

Quantização/artefato: MLX-4bit · MLX204.0 GB · Public file metadata
Capacidade no Mac selecionado: -195.99GB restantes após a reserva
Evidência de tempo de execução: Relato da comunidademlx
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public file metadata204.0 GB · exact public file metadata
Fonte do runtimemlx · Relato da comunidade
Evidência da comunidade (2)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto.
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. Limitação: 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. Limitação: 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.

Use no consultor gratuito

GLM · Limite

GLM-5.3 744B-A40B

GLM-5.3 · 744B total / 40B active

Não cabe

Quantização/artefato: MLX-4bit · MLX446.4 GB · Public model source; exact file not registered
Capacidade no Mac selecionado: -438.4GB restantes após a reserva
Evidência de tempo de execução: Desconhecidollama.cpp
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public model source; exact file not registered446.4 GB · parameter planning estimate
Fonte do runtimellama.cpp · Desconhecido
Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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.

Use no consultor gratuito

Google Gemma · Pequeno

Gemma 4 E2B

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

Com margem

Quantização/artefato: Q4_0 · QAT3.35 GB · Public file metadata
Capacidade no Mac selecionado: 4.65GB restantes após a reserva
Evidência de tempo de execução: Relato da comunidadellama.cpp
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public file metadata3.35 GB · exact public file metadata
Fonte do runtimellama.cpp · Relato da comunidade
Evidência da comunidade (1)Os relatórios públicos são apenas sinais e nunca transformam esta página num teste do projeto.
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. Limitação: 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.

Use no consultor gratuito

Google Gemma · Principal

Gemma 4 12B Unified

Gemma 4 · 11.95B dense

Ajustado

Quantização/artefato: Q4_0 · QAT6.98 GB · Public file metadata
Capacidade no Mac selecionado: 1.02GB restantes após a reserva
Evidência de tempo de execução: Desconhecidollama.cpp
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public file metadata6.98 GB · exact public file metadata
Fonte do runtimellama.cpp · Desconhecido
Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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.

Use no consultor gratuito

Google Gemma · Limite

Gemma 4 26B-A4B

Gemma 4 · 25.2B total / 3.8B active

Não cabe

Quantização/artefato: Q4_0 · QAT14.4 GB · Public file metadata
Capacidade no Mac selecionado: -6.44GB restantes após a reserva
Evidência de tempo de execução: Relato da comunidadellama.cpp
Confiança: highVerificado 2026-08-28

Fontes e limites

Fonte oficial do modeloModelo family, parameters, and release context · Verificado 2026-08-28
Public file metadata14.4 GB · exact public file metadata
Fonte do runtimellama.cpp · Relato da comunidade
Nesta atualização não foi captado nenhum relato fiável de primeira mão sobre Mac.A capacidade continua a ser uma estimativa de planeamento baseada em metadados públicos; a ausência de provas não valida o 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.

Use no consultor gratuito

Evidência de capacidade

A capacidade de fluxo de trabalho é uma questão separada

As janelas de contexto são metadados de modelo. Simultaneidade, chamada de ferramenta, confiabilidade JSON, visão e comportamento sustentado dependem do modelo exato, do tempo de execução, do cliente, da pressão de memória e da tarefa. Uma célula positiva ainda não é uma garantia para o seu fluxo de trabalho.

Modelo	Contexto	Concorrência	Chamadas de ferramentas	JSON	Visão	Uso contínuo
Qwen3.8 27BPequeno	Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Suporte oficialThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	DesconhecidoVision support is not assumed from family branding or an unrelated model variant.	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Qwen3.8-Flash-Next 180B (6B active)Principal	Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Suporte oficialThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	Suporte oficialThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Qwen3.8 2.4T-A95BLimite	Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Suporte oficialThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	DesconhecidoVision support is not assumed from family branding or an unrelated model variant.	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
DeepSeek-V4-Flash 284B-A13BPrincipal	Suporte oficialPublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Suporte oficialThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	DesconhecidoVision support is not assumed from family branding or an unrelated model variant.	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
DeepSeek-V4-Pro 1.6T-A49BLimite	Suporte oficialPublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Suporte oficialThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	DesconhecidoVision support is not assumed from family branding or an unrelated model variant.	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
GLM-5.3-Flash 320B-A18BPrincipal	Suporte oficialPublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Suporte oficialThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte	Benchmark públicoPublic GLM evaluations cover agentic or structured tasks, but they do not establish JSON validity in every client.Fonte	Suporte oficialThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
GLM-5.3 744B-A40BLimite	Suporte oficialPublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	Suporte oficialThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	DesconhecidoVision support is not assumed from family branding or an unrelated model variant.	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Gemma 4 E2BPequeno	Suporte oficialPublic model metadata lists 131,072 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	DesconhecidoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	Suporte oficialThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Gemma 4 12B UnifiedPrincipal	Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	DesconhecidoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	DesconhecidoVision support is not assumed from family branding or an unrelated model variant.	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.
Gemma 4 26B-A4BLimite	Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte	Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte	DesconhecidoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.	DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.	DesconhecidoVision support is not assumed from family branding or an unrelated model variant.	DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Qwen3.8 27BPequeno

Contexto: Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: Suporte oficialThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: DesconhecidoVision support is not assumed from family branding or an unrelated model variant.
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

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

Contexto: Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: Suporte oficialThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: Suporte oficialThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Qwen3.8 2.4T-A95BLimite

Contexto: Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: Suporte oficialThe current Qwen material documents tool or agent paths. The exact model template and runtime still need a workflow check.Fonte
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: DesconhecidoVision support is not assumed from family branding or an unrelated model variant.
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

DeepSeek-V4-Flash 284B-A13BPrincipal

Contexto: Suporte oficialPublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: Suporte oficialThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: DesconhecidoVision support is not assumed from family branding or an unrelated model variant.
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

DeepSeek-V4-Pro 1.6T-A49BLimite

Contexto: Suporte oficialPublic model metadata lists 1,048,576 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: Suporte oficialThe official DeepSeek V4 material documents tool use in the model family. The exact local template and runtime still need a workflow check.Fonte
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: DesconhecidoVision support is not assumed from family branding or an unrelated model variant.
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

GLM-5.3-Flash 320B-A18BPrincipal

Contexto: Suporte oficialPublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: Suporte oficialThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte
JSON: Benchmark públicoPublic GLM evaluations cover agentic or structured tasks, but they do not establish JSON validity in every client.Fonte
Visão: Suporte oficialThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

GLM-5.3 744B-A40BLimite

Contexto: Suporte oficialPublic model metadata lists 1,000,000 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: Suporte oficialThe official GLM-5.3 material documents function calling and agent capabilities. It is not a Mac runtime integration benchmark.Fonte
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: DesconhecidoVision support is not assumed from family branding or an unrelated model variant.
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Gemma 4 E2BPequeno

Contexto: Suporte oficialPublic model metadata lists 131,072 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: DesconhecidoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: Suporte oficialThe current model documentation describes multimodal inputs for this representative; the selected Mac runtime path still needs checking.Fonte
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Gemma 4 12B UnifiedPrincipal

Contexto: Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: DesconhecidoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: DesconhecidoVision support is not assumed from family branding or an unrelated model variant.
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Gemma 4 26B-A4BLimite

Contexto: Suporte oficialPublic model metadata lists 262,144 tokens. This is not proof that the selected Mac can hold that context with the weights.Fonte
Concorrência: Inferência do sistemaConcurrency is inferred only from memory budget and runtime architecture. No apples-to-apples public concurrency benchmark is attached.Fonte
Chamadas de ferramentas: DesconhecidoNo model-specific public tool-calling result is strong enough for a positive claim in this snapshot.
JSON: DesconhecidoStructured JSON behavior depends on the exact model template, runtime parser, and client; no exact public result is registered here.
Visão: DesconhecidoVision support is not assumed from family branding or an unrelated model variant.
Uso contínuo: DesconhecidoNo public, comparable sustained Apple Silicon run is registered for this model and Mac configuration.

Como ler os rótulos

Suporte oficial A documentação oficial do modelo/runtime descreve o caminho de capacidade; o comportamento exato do cliente ainda precisa de verificação.
Benchmark público Existe uma avaliação publicada, mas não é um benchmark de desempenho ou fluxo de trabalho no Mac.
Relato da comunidade Relatórios públicos ou artefactos da comunidade são sinais, não testes do projeto nem garantias do fornecedor.
Inferência do sistema Inferência limitada pela arquitetura de capacidade/runtime, não um resultado observado.
Desconhecido Não foi recolhida evidência pública fiável para esta afirmação exata.
Verificado pelo projeto Só é verificada a combinação exata de modelo, quantização, runtime e máquina no registo de testes do projeto.

Traga sua carga de trabalho real

Transforme a evidência pública em uma decisão pessoal.

O Advisor gratuito adiciona seu Mac atual, contexto, simultaneidade, orçamento, compatibilidade e requisitos offline. Pode dizer “precisa de verificação” quando as evidências públicas não são suficientes.

Execute o consultor gratuito

Perguntas frequentes

Verifique os limites de prova antes de comprar.

Este é um benchmark Keep ou Upgrade?

Não. É uma ferramenta de síntese e planeamento de evidências públicas. Apenas a combinação exata verificada nos testes de fumaça (smoke tests) do projeto será exibida como verificada.

Como você calcula a capacidade do seu Mac?

Use o tamanho do arquivo público selecionado se o arquivo exato ou o total do arquivo dividido estiver disponível; caso contrário, use a estimativa de planejamento explícita. O orçamento modelo reserva espaço para macOS, tempo de execução e contexto.

O MoE requer menos memória se for um parâmetro válido?

Não. Os pesos totais armazenados são a condição de capacidade. Os parâmetros válidos descrevem apenas a quantidade de computação por token.

A duração oficial do contexto funciona em qualquer Mac?

Não. O comprimento do contexto são metadados do modelo. Depois de peso, tempo de execução, sistema e outros aplicativos, você também precisa ter memória suficiente no Mac selecionado.

Pode comprar um Mac apenas olhando para a tabela?

Depois de restringir seu escopo de planejamento com esta matriz, insira seu Mac atual, contexto, simultaneidade, clientes, orçamento, compatibilidade e requisitos off-line em nosso consultor gratuito. A evidência pública não é uma garantia de compra.

trazer o trabalho real

Use evidências públicas para informar suas decisões de compra.

O Free Advisor adiciona seu Mac atual, contexto, simultaneidade, orçamento, compatibilidade, requisitos off-line e indica que a verificação é necessária se não houver evidências públicas suficientes.

Comece o consultor gratuito

Keep or Upgrade é uma ferramenta independente; a Apple não patrocina nem recomenda esta página.