Mistral launches Large 4, a trillion-parameter model with a lower price tag

Mistral has launched Large 4, a trillion-parameter AI model that activates 49 billion parameters per answer and is priced well below several leading rivals. The launch gives developers a substantial new open-weight model to assess, though Mistral says its weights will not be available until the end of October.

Mistral AI Watch analysis

What happened

Large 4 uses a mixture-of-experts design: it has one trillion parameters in total, but routes each question to a smaller active subset. Mistral’s previous flagship, Large 3, activated 41 billion of its 675 billion parameters per answer.

Decrypt reports that Large 4 costs $1.36 per million input tokens and $4.18 per million output tokens. In a blind human coding evaluation by Surge AI, it ranked second of five models, scoring 3.74 out of 5. On DeepSWE 1.1, Mistral reports a score of 62, ahead of GLM-5.3 and DeepSeek V4 Pro, but behind Kimi K3. Mistral says it plans to release the model’s weights by the end of October.

Why it matters

The combination of a large model, a relatively small active parameter count and lower listed token prices makes Large 4 an interesting option for teams weighing performance against running costs. The benchmark picture is mixed rather than triumphant: it places well in some comparisons, but the reported coding evaluation does not put it first.

Open weights could make the claims easier for outside developers to test, but that practical check is still pending. For now, the reported prices and results are useful starting points, not a guarantee of what any particular workload will cost or achieve.

Our read

Large 4 looks like a serious new contender, not a leaderboard coronation. The most useful detail is the balance between its total size and the 49 billion parameters active per answer; the most useful next step is seeing how it behaves when developers can actually run it. If you are comparing models now, treat Mistral’s reported results as a shortlist clue, not a procurement verdict.

What to watch

• Whether Mistral releases the weights by the end of October.
• How independent tests compare Large 4 with other open-weight models.
• Whether its listed prices translate into lower costs on real workloads.

Discussion spark: Would you choose a lower-priced model with promising but mixed benchmark results, or wait for independent testing before putting it into a real workflow?

not affiliated with or endorsed by Mistral AI