AI & Automation
Mistral Large 3 Released: 128K Context and Stronger Coding Benchmarks
SedulousWebVia Mistral AI
Mistral AI released Large 3 with a 128K token context window and improved coding scores on HumanEval, positioning it as a GPT-4o competitor at lower API cost.
The Release
Mistral AI shipped Mistral Large 3 via the La Plateforme API. The model doubles the context window to 128K tokens and posts a 72.3% score on HumanEval — up from 64.1% in Large 2.
Benchmark Highlights
- HumanEval: 72.3% (up from 64.1%)
- MMLU: 88.2%
- Context window: 128K tokens
Pricing
Input: $2 per million tokens. Output: $6 per million tokens — roughly 30% cheaper than GPT-4o at comparable benchmarks.
Practical Takeaway
If you run mistral-large-latest in production, no code changes are needed — the alias auto-upgrades. Test your longest prompts before relying on it in prod.