From model promises to work we can check
Researched retrospectively on 13 October 2026. Mistral opens a model preview. OpenAI shows ways to evaluate multi-step office work and mathematical results.
Mistral Large 4 enters public preview
On 6 October, Mistral released an API preview of Large 4. The model accepts different kinds of input, including text and images. Model weights were announced for the end of October; they had not been released with this announcement. Mistral reports progress in coding and business tasks. These are vendor claims, not Applied by Hand test results.
What it means: Our interpretation: An API preview differs from a model you can run yourself.
Example: A possible approach: A team tests the preview on its own documents before deciding.
- Mistral · Introducing Mistral Large 4 · 6 October 2026
Also worth knowing
Contract workflows become an agent evaluation
OpenAI and Ironclad presented research tasks for contracting software. An agent might configure approval rules and test different requests. Scoring measures specific criteria. A higher average score does not mean error-free contract work.
What it means: Our interpretation: The complete process matters more than individual clicks.
Example: For example, an expensive purchase needs finance approval; a cheaper one does not.
- OpenAI · Advancing computer use with Ironclad · 6 October 2026
New mathematical results become open to scrutiny
OpenAI published mathematical results from an internal model in a GitHub repository. Many proofs include versions in Lean, a language for checking mathematical steps. More formalizations are planned. Publication alone does not validate every result.
What it means: Our interpretation: Checkable steps matter more than impressive claims.
Example: A possible use: A researcher checks a formal proof and examines its assumptions.
- OpenAI · Sharing AI progress in mathematics · 6 October 2026