Mistral Large 4 Weights Miss October 27 — 969 on the Wait List, License Still Unknown
October 27 arrived without Mistral Large 4 weights. The Hugging Face placeholder for the model, fetched at 14:46Z Tuesday, shows the same state it held Monday: an “Upcoming release” badge, a countdown to October 31 at 15:00Z, a single planned artifact, and no downloadable model files. The wait list grew from 813 on October 14 to 969 by Tuesday morning.
Reuters reported October 27 as the weight release date on October 6, but Mistral never published that date directly. The company’s own announcement, unchanged since October 7, commits only to “Weights drop end of this month” — a sentence broad enough to absorb a four-day slip to October 31.
What the Model Is
Mistral Large 4 launched as an API preview on October 6. It is a mixture-of-experts model with approximately one trillion total parameters and 49 to 52 billion active parameters per token — sources conflict on the precise active count. It is natively multimodal. Preview API pricing is $1.36 per million input tokens and $4.18 per million output tokens, with a launch discount that halves both figures while it runs.
Vendor-reported benchmarks from the October 6 launch:
| Benchmark | Score | Context |
|---|---|---|
| DeepSWE v1.1 | 62% | Claimed to beat GLM 5.3 |
| Finance Agent v2 | 54.7 | Above GPT-6 Astra (53.5) |
| Finch / FinWorkBench | 67% | Financial agent tasks |
Independent replication of these figures has not been published.
The License Question
The license is the unresolved issue with material consequences. No license file is attached to the Hugging Face repository. Mistral Large 3 shipped under Apache 2.0, which permits unlimited commercial use including cloud API resale. Large 4 may not.
Mistral’s existing open-weight models include releases under a Modified MIT variant that caps free commercial use at companies generating less than $20 million in monthly revenue. Organizations above that threshold must negotiate a separate commercial agreement with Mistral or route traffic through the Mistral hosted API. At one trillion parameters, almost no enterprise will self-host anyway — but inference providers, cloud platforms, and model aggregators that serve the model as an API would fall under the commercial threshold quickly.
If Large 4 ships Apache 2.0, cloud inference providers can offer it freely alongside each other. If it ships under the revenue-capped terms, the model’s open-weight status becomes largely symbolic at commercial scale.
Mistral has not published the license terms on its site, in its announcement post, or on the Hugging Face card. The placeholder field for the license on the repository remains empty.
What to Watch
The October 31, 15:00Z timestamp on the HuggingFace card is labeled as “approximate ETA” with a tooltip warning it may still change. The October 6 Mistral announcement page has not been updated since October 7. Mistral’s blog index, as of Tuesday, shows no new post beyond that October 6 entry.
The practical uncertainty for organizations planning deployments: they cannot finalize integration plans until the license terms are public. The model’s benchmark positioning — above GPT-6.1 Sol on Finance Agent, claimed parity with GLM 5.3 on code tasks — makes that delay commercially significant for the financial services and enterprise coding use cases Mistral targeted in the October 6 launch.