Mistral Medium 3.5
A mid-scale language model released by Mistral, positioned as a predecessor to Mistral Large 4. The model offers moderate capability on agent-oriented tasks, sitting within Mistral's tiered product lineup between their smaller and larger model offerings.
Timeline (last 90 days)
The M5 Ultra GPU neural accelerator improved the TTFT of Qwen 3.8 and Mistral Medium 3.5 by up to 4.6 times.
On a single Mac Studio with 512GB of memory, the 128B Mistral Medium 3.5 (FP8) and the 27B Qwen 3.8 (BF16) occupy about 158GB and 68GB of memory respectively at a 65536-token context.
Dense models (Qwen 3.8, Mistral Medium 3.5) perform better than MoE models (Muse Spark Open) on multi-machine scaling, as the latter's expert outputs need to be normalized separately, introducing extra prefill communication overhead.
Mistral Large 4's AA score is 38, a significant improvement over the previous generation Mistral Medium 3.5's score of 14.
All Facts (7)
Mistral Medium 3.5在HumanEval基准测试中通过率为77.6%
90%UnverifiedMistral Medium 3.5采用修改版MIT许可的开放权重模型
90%UnverifiedMistral Medium 3.5是Mistral于2025年4月发布的旗舰模型，专为智能体和编码工作场景优化
90%UnverifiedDense models (Qwen 3.8, Mistral Medium 3.5) perform better than MoE models (Muse Spark Open) on multi-machine scaling, as the latter's expert outputs need to be normalized separately, introducing extra prefill communication overhead.
50%UnverifiedThe M5 Ultra GPU neural accelerator improved the TTFT of Qwen 3.8 and Mistral Medium 3.5 by up to 4.6 times.
50%UnverifiedOn a single Mac Studio with 512GB of memory, the 128B Mistral Medium 3.5 (FP8) and the 27B Qwen 3.8 (BF16) occupy about 158GB and 68GB of memory respectively at a 65536-token context.
50%UnverifiedMistral Large 4's AA score is 38, a significant improvement over the previous generation Mistral Medium 3.5's score of 14.
50%