Mistral Large 3 Reasoning

Mistral announced a reasoning-tuned variant of Mistral Large 3 as "coming soon" during the Mistral 3 launch on December 2, 2025 — extended chain-of-thought on the 675B MoEA model architecture that splits the model into many smaller specialized "expert" networks, only activating a handful per input rather than running the whole model every time. The practical effect: you get the knowledge capacity of a big model with the compute cost of a much smaller one. Mistral Large 3 and Mistral Small 4 are both MoE models. architecture (41B active), expected to match Mistral Large 3 Instruct. License posture is assumed to match the broader Large 3 family (Apache 2.0), unverified until release. As of April 24, 2026 — nearly five months later — no release date has been confirmed and no public checkpointA specific saved version of a model at a particular point in training. When a creator releases "Llama 3.1 8B Instruct," they're releasing a checkpoint — a frozen snapshot of the model as it existed at the end of training. Most models ship only a single public checkpoint; some creators release multiple (base, instruct, reasoning variants of the same underlying model). has appeared on Hugging Face or Mistral's API.

It would be the flagship-tier reasoning counterpart to Mistral Large 3 Instruct — a major open-weightA model where the trained weights are freely downloadable — you can run it yourself without contacting the creator. Llama, Mistral, Qwen, and Gemma are open-weight. Open-weight does not mean open-source: the training data and code often stay private. The license still governs what you can do with the weights, including whether you can use them commercially. reasoning modelA model trained to "think through" problems step by step before answering, often by producing internal reasoning that's either shown or hidden from the user. Reasoning models trade speed for accuracy on hard problems — they're slower and more expensive per answer, but markedly better at math, logic, and complex analysis. OpenAI's o1 series and Mistral's Magistral are reasoning models. if it ships.

Announced but never shipped: no confirmed release date and no public checkpointA specific saved version of a model at a particular point in training. When a creator releases "Llama 3.1 8B Instruct," they're releasing a checkpoint — a frozen snapshot of the model as it existed at the end of training. Most models ship only a single public checkpoint; some creators release multiple (base, instruct, reasoning variants of the same underlying model).. It may also be functionally stale — Mistral's Small 4 (March 2026) folded reasoning into a single configurable checkpoint (a reasoning_effort parameter), so a future Large release could be "Mistral Large 4" with built-in configurable reasoning rather than a standalone "Large 3 Reasoning" variant.

This is a watchlist entry, not a shipped model. Mistral announced a reasoning variant of Mistral Large 3 as "coming soon" during the Mistral 3 launch on December 2, 2025. As of April 24, 2026 — nearly five months later — no release date has been confirmed and no public checkpointA specific saved version of a model at a particular point in training. When a creator releases "Llama 3.1 8B Instruct," they're releasing a checkpoint — a frozen snapshot of the model as it existed at the end of training. Most models ship only a single public checkpoint; some creators release multiple (base, instruct, reasoning variants of the same underlying model). has appeared on Hugging Face or Mistral's API.

There's a plausible reason it hasn't shipped. In March 2026, Mistral released Mistral Small 4 with a new architectural approach: a single model that handles instruction-following, multimodalA model that can handle more than one type of input or output — typically text plus images, sometimes plus audio or video. "GPT-4 Vision" and "Llama 3.2 11B Vision" are multimodal models that accept both text and images. A text-only model is called "unimodal" but nobody uses that term; text-only is the assumed default. vision, agentic coding, AND reasoning via a configurable reasoning_effort parameter — effectively absorbing the previous standalone Magistral reasoning line into a unified checkpoint. If Mistral is applying the same pattern at the Large tier, the future release might not be "Mistral Large 3 Reasoning" as a separate variant but rather "Mistral Large 4" with unified configurable reasoning built in. That would make this announcement functionally stale even though it hasn't been formally retracted.

Recommended catalog review cadence: monthly. Check for (a) a standalone Large 3 Reasoning release, (b) a Mistral Large 4 release that supersedes this announcement, or (c) formal retraction. If release hasn't happened by end of Q3 2026, consider reclassifying to archived with a note explaining the non-delivery and the likely Small-4-unified-architecture explanation.