New models & updates
2026-10-15
RunItLocal launches
24 GPUs and Macs, 21 open-weight models, 504 hardware × model answer pages, and a memory calculator.
2026-10-15
More realistic MoE speed estimates
Mixture-of-experts models (gpt-oss, Qwen3 30B-A3B / 235B-A22B, Llama 4 Scout, DeepSeek V3) now include routing and shared-weight overhead. Quants that only fit at 4K context are now flagged.
All tracked models (21)
- Llama 3.1 8B Instruct Meta
- Llama 3.3 70B Instruct Meta
- Llama 3.2 3B Instruct Meta
- Llama 4 Scout (17B active, 16 experts) Meta
- Qwen3 8B Alibaba Qwen
- Qwen3 14B Alibaba Qwen
- Qwen3 32B Alibaba Qwen
- Qwen3 30B-A3B (MoE) Alibaba Qwen
- Qwen3 235B-A22B (MoE) Alibaba Qwen
- Qwen2.5 Coder 32B Instruct Alibaba Qwen
- Gemma 3 4B Google
- Gemma 3 12B Google
- Gemma 3 27B Google
- Mistral Small 3.1 24B Mistral AI
- Mistral Nemo 12B Mistral AI
- Phi-4 14B Microsoft
- DeepSeek R1 Distill Qwen 32B DeepSeek
- DeepSeek R1 Distill Llama 70B DeepSeek
- gpt-oss-20b (MoE) OpenAI
- gpt-oss-120b (MoE) OpenAI
- DeepSeek V3 / R1 671B (MoE) DeepSeek
Get new models & benchmarks in your inbox
No spam. Unsubscribe anytime by replying to any email.