Mistral Medium 3.5 (Mistral AI)

Mistral's April 29, 2026 release bundled Mistral Medium 3.5 with remote coding agents in Vibe and Work mode in Le Chat. The 128B dense model with a 256k context window ships under a modified MIT license, supports configurable reasoning effort per request, and self-hosts on as few as four GPUs. The notable differentiator is that weights are downloadable, unlike competing cloud coding agents from Cursor, Claude Code, or GitHub Copilot. Vibe's remote agents run coding sessions in parallel cloud sandboxes, can spawn from the CLI or Le Chat, and allow local sessions to be teleported to the cloud mid-task. Integrations span GitHub for pull requests, Linear and Jira for issues, Sentry for incidents, and Slack and Teams for notifications. API pricing is $1.5 per million input tokens and $7.5 per million output tokens, with the model scoring 77.6% on SWE-Bench Verified.

Mistral Medium 3.5 is documented on AI Evals as a Mistral AI open-weight model released on 29 April 2026 with a 262k-token context window and 80k-token output. It is a 128B-parameter dense model served locally by Ollama from an 80GB download, supporting vision input, tool calling, reasoning mode, and text/image inputs. Listed API pricing is $1.50 per million input tokens and $7.50 per million output tokens. Across six benchmarks on AI Evals, the model placed 13th–31st among 32 listed models, including 66.40% on SWE-bench Verified, 75.34% on MMLU Pro, and 34.85% on GPQA Diamond. It took no first-place spot among the independently published columns.

Mistral AI announced Mistral Medium 3.5, a 128B dense flagship model that merges instruction-following, reasoning, and coding capabilities within a 256k context window. The model ships as open weights under a modified MIT license on Hugging Face and can be self-hosted on as few as four GPUs, making local deployment practical for organizations with sufficient hardware. It replaces Devstral 2 as the default model in Vibe CLI and Le Chat. Mistral Medium 3.5 posts 77.6% on SWE-Bench Verified, ahead of Devstral 2 and Qwen3.5, and scores 91.4 on τ³-Telecom, signaling strong agentic coding and telecom-tool performance. API pricing is set at $1.5 per million input tokens and $7.5 per million output tokens, positioning it as a cost-efficient alternative for frontier tasks. Availability spans Mistral Pro, Team, and Enterprise plans, plus NVIDIA NIM and build.nvidia.com.