willitrun·ai

Will It Run AI · 计算器

告诉我们你有什么硬件、想做什么任务，我们会为你排出合适的本地模型。

从你的硬件和工作负载出发，基于适配度、速度和运行时支持生成推荐列表，而不是靠猜测通用模型列表或基准截图。

从你的硬件开始查看排名机制

实时目录快照：196 hardware profiles, 380 models, 24 runtimes。这使计算器与当前目录保持同步，而非依赖静态基准列表。

正在评估

RTX 4070 12GB

工作负载

Coding

运行时

llama.cpp

Operating mode

Balanced

输入

选择要测试的硬件、运行时和工作负载。

如果检测到的硬件正确就直接使用，不正确则手动修改，然后重新运行排名来比较实际的本地 AI 选项。

Browser detection

Collecting GPU metadata…

Awaiting detection

Update the hardware or workload and recalculate to refresh the ranking.

1. 适配

内存适配度和余量决定了模型在所选硬件上是否可行。

2. 工作负载

评分会奖励匹配所选任务的模型，并在有更新的专用版本时惩罚过时或旧版模型系列。

3. 速度

解码吞吐量和 TTFT 确保推荐列表适用于实际使用，而不仅是理论上可以运行。

Qwen

Alibaba

Qwen 3.5 9B

前沿发布于 Jun 2025Hugging FaceOllamaLM Studio

为何胜出

Qwen 3.5 9B is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 1 名
SRunsMEASURED

评分

122.0

适配状态

Runs well

适配：Runs well，安全上下文 32K。

运行时支持：native，通过 GGUF 在 cuda-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

72 tok/s

安全上下文

32K

官方上下文

131K

支持

native

首 Token 延迟

2616 ms

权重：5.5 GB

KV 缓存：2.2 GB

后端：cuda-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 122.0 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

CodeGeeX

Tsinghua/Zhipu

CodeGeeX 4 9B

当前发布于 Jul 2024Hugging FaceOllama

为何胜出

CodeGeeX 4 9B is a specialized fit for Coding. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 2 名
ARunsEST.

评分

114.6

适配状态

Runs well

适配：Runs well，安全上下文 116K。

运行时支持：native，通过 GGUF 在 cpu-gpu-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

75.3 tok/s

安全上下文

116K

官方上下文

131K

支持

native

首 Token 延迟

2571 ms

权重：5.5 GB

KV 缓存：0.6 GB

后端：cpu-gpu-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 114.6 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Gemma

Google

Gemma 4 E4B

前沿发布于 Apr 2026Hugging FaceOllamaLM Studio

为何胜出

Gemma 4 E4B is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 3 名
ARunsEST.

评分

110.2

适配状态

Runs well

适配：Runs well，安全上下文 63K。

运行时支持：native，通过 GGUF 在 cpu-gpu-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

63.1 tok/s

安全上下文

63K

官方上下文

128K

支持

native

首 Token 延迟

3068 ms

权重：4.9 GB

KV 缓存：1.3 GB

后端：cpu-gpu-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 110.2 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Codestral

Mistral AI

Codestral Mamba 7B

当前发布于 Jul 2024Hugging FaceOllama

为何胜出

Codestral Mamba 7B is a specialized fit for Coding. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 4 名
ARunsEST.

评分

107.2

适配状态

Runs well

适配：Runs well，安全上下文 184K。

运行时支持：native，通过 GGUF 在 cuda-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

98 tok/s

安全上下文

184K

官方上下文

262K

支持

native

首 Token 延迟

1976 ms

权重：4.3 GB

KV 缓存：0.5 GB

后端：cuda-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 107.2 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Yi

01.AI

Yi Coder 9B

当前发布于 Sep 2024Hugging FaceOllamaLM Studio

为何胜出

Yi Coder 9B is a specialized fit for Coding. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 5 名
BRunsEST.

评分

106.6

适配状态

Runs well

适配：Runs well，安全上下文 48K。

运行时支持：native，通过 GGUF 在 cuda-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

74.9 tok/s

安全上下文

48K

官方上下文

131K

支持

native

首 Token 延迟

2586 ms

权重：5.5 GB

KV 缓存：1.5 GB

后端：cuda-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 106.6 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Granite

IBM

Granite 4.1 8B

当前发布于 Apr 2026Hugging FaceOllama

为何胜出

Granite 4.1 8B is a specialized fit for Coding. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 6 名
ARunsEST.

评分

102.3

适配状态

Runs well

适配：Runs well，安全上下文 33K。

运行时支持：native，通过 GGUF 在 cpu-gpu-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

83.3 tok/s

安全上下文

33K

官方上下文

131K

支持

native

首 Token 延迟

2325 ms

权重：4.9 GB

KV 缓存：2.4 GB

后端：cpu-gpu-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 102.3 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Qwen

Alibaba

Qwen 2.5 Coder 7B

当前发布于 Sep 2024Hugging FaceOllamaLM Studio

为何胜出

Qwen 2.5 Coder 7B is a specialized fit for Coding. It sits in the middle of the current generation mix. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 7 名
ARunsEST.

评分

101.0

适配状态

Runs well

适配：Runs well，安全上下文 105K。

运行时支持：native，通过 GGUF 在 cuda-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

96.1 tok/s

安全上下文

105K

官方上下文

131K

支持

native

首 Token 延迟

2014 ms

权重：4.3 GB

KV 缓存：0.9 GB

后端：cuda-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 101.0 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Qwen

Alibaba

Qwen 3 8B

前沿发布于 Apr 2025Hugging FaceOllamaLM Studio

为何胜出

Qwen 3 8B is viable for Coding, but is not the most specialized choice. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 8 名
SRunsEST.

评分

99.6

适配状态

Runs well

适配：Runs well，安全上下文 37K。

运行时支持：native，通过 GGUF 在 cpu-gpu-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

83.3 tok/s

安全上下文

37K

官方上下文

131K

支持

native

首 Token 延迟

2325 ms

权重：4.9 GB

KV 缓存：2.2 GB

后端：cpu-gpu-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 99.6 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Nemotron

NVIDIA

Nemotron Nano 9B v2

前沿发布于 Jun 2025Hugging FaceOllamaLM Studio

为何胜出

Nemotron Nano 9B v2 is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It should run, but memory headroom will be limited. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.

Capacity: Tight · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Good · Bottleneck: Balanced

第 9 名
ATightEST.

评分

99.4

适配状态

Tight fit

适配：Tight fit，安全上下文 29K。

运行时支持：native，通过 GGUF 在 cuda-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q4-k-m

解码

74 tok/s

安全上下文

29K

官方上下文

131K

支持

native

首 Token 延迟

2616 ms

权重：5.5 GB

KV 缓存：2.4 GB

后端：cuda-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 99.4 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

Qwen

Alibaba

Qwen 3.5 4B

前沿发布于 Jun 2025Hugging FaceOllamaLM Studio

为何胜出

Qwen 3.5 4B is a specialized fit for Coding. It is a recent-generation family, which helps on current local SOTA workloads. It fits natively with comfortable headroom. Context coverage stays within the requested workload envelope. Known distribution channels: huggingface, ollama, lm-studio.

Capacity: Roomy · Bandwidth: Medium · Stack: Standard

Interactive: Good · Light API: Great · Bottleneck: Balanced

第 10 名
SRunsEST.

评分

93.6

适配状态

Runs well

适配：Runs well，安全上下文 48K。

运行时支持：native，通过 GGUF 在 cpu-gpu-local 上运行。

运行时

llama.cpp

制品

GGUF

量化

q6-k

解码

56 tok/s

安全上下文

48K

官方上下文

131K

支持

native

首 Token 延迟

3457 ms

权重：3.3 GB

KV 缓存：2.2 GB

后端：cpu-gpu-local

Current limits

This setup is broadly balanced for this model.

No major red flags

This recommendation has enough memory headroom and acceptable estimated speed for the selected workload.

Best next improvements

评分 93.6 综合了工作负载匹配、目录新鲜度、适配安全性、上下文覆盖、制品选择、内存利用率、吞吐量和延迟。

全部 380 个模型

Full compatibility grid for RTX 4070 12GB

246 models fit · 9 excellent · 38 great

Grade
Model
Params
Tasks
Q4 VRAM
Decode
Context
Memory
Fit