Mistral Large 4
mistral · Released Oct 2026
Conditional: Mistral Large 4 is a public preview, not a finished release, since Mistral says its reinforcement learning run is still in flight and promises weights "by the end of the month". It is a mixture-of-experts model of about 1 trillion parameters, 52B active, with a 1M context on its docs page and a $1.36 input, $4.18 output list price per 1M tokens. Most benchmarks are Mistral's own. Trial it now if you want an open-weight-bound model, but wait for weights and licence before committing.
Is it right for you?
Good for
- Trying out a 1.05T-parameter, 52B-active mixture-of-experts model with text and image input through Mistral Studio while it is in public preview
- Long-document and agent work that needs a 1M-token context (per the docs model page)
- Teams that want an open-weight path: Mistral says weights follow by the end of the month
- Apps that need function calling, structured outputs, document QnA and batching, all listed on the docs model page
Not good for
- Production commitments: this is a public preview and Mistral says the RL run behind it is still in flight
- Self-hosting today: no weights are out and no licence was published on the pages we read
- Decisions that need independent evaluations: most scores are Mistral's own, and Mistral ran the blind human coding evaluation with Surge AI
- Budgeting off the docs page alone, which shows a sale rate that differs from the announcement's list rate
How it performs by task
Agentic coding
DeepSWE 1.1 61.7% and Terminal Bench 4.0 28.3% are Mistral's numbers; Mistral's blind human evaluation with Surge AI ranks it second at 3.74/5 behind Claude Opus 5 at 4.22; untested by us
Cybersecurity
Mistral reports 93% of the 40 Cybench challenges and 82% on one test of the AA Cyber Index (reproduce a real open-source vulnerability, then patch it); both are Mistral's numbers, and the 82% is one test, not the index overall; untested by us
Pricing
Input
$1.36 / 1M tokens
Output
$4.18 / 1M tokens
Context
1M ctx; preview; sale $0.68/$2.09 on docs
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| DeepSWE 1.1 (Mistral-reported) | 61.7% | Source |
| Terminal Bench 4.0 (Mistral-reported) | 28.3% | Source |
| AutomationBench (Mistral-reported) | 59.9% | Source |
| Cybench (Mistral-reported) | 93% | Source |
| Surge AI blind human coding evaluation, 1-5 scale (run by Mistral with Surge AI; Claude Opus 5 scored 4.22) | 3.74 | Source |
No verdict changes yet
The clock starts day one. Changes land here as our verdict evolves.
Sources
Verification log
- Pricing· No changes
Automated agent