Codestral 22B is 87% cheaper than Llama 3.1 405B (Together AI) on blended 3:1 token pricing. Codestral 22B costs $0.30/1M in and $0.90/1M out, while Llama 3.1 405B (Together AI) costs $3.50/1M in and $3.50/1M out.
Codestral 22B vs Llama 3.1 405B (Together AI)
Compare verified input/output token rates, prompt caching discounts, latency tiers, and context limits between Mistral AI and Together AI (Meta).
| Metric / Feature | Codestral 22B (Mistral AI) | Llama 3.1 405B (Together AI) (Together AI (Meta)) |
|---|---|---|
| Input Price ($ / 1M Tokens) | $0.30 | $3.50 |
| Output Price ($ / 1M Tokens) | $0.90 | $3.50 |
| Prompt Caching Input Rate | Not Available | Not Available |
| Blended 3:1 Rate (Production) | $0.450 / 1M | $3.500 / 1M |
| Max Context Window | 128k | 128k |
| Multimodal Vision | ❌ Text Only | ❌ Text Only |
| MMLU Benchmark | 81.1% | 88.6% |
| Throughput & Latency | Ultra-Fast (~80 t/s) | Standard (~30 t/s) |
| Best For | Fill-in-the-middle code completion, automated unit tests, IDE integration | Synthetic data generation, model distillation, enterprise on-premise benchmarking |
When to Choose Codestral 22B
Choose Codestral 22B if your workload requires fill-in-the-middle code completion, automated unit tests, ide integration. Ideal for teams needing Mistral AI's infrastructure and ecosystem tooling.
When to Choose Llama 3.1 405B (Together AI)
Choose Llama 3.1 405B (Together AI) if your primary objective is synthetic data generation, model distillation, enterprise on-premise benchmarking. Ideal for scaling high-throughput pipelines with Together AI (Meta).
Want to compare Codestral 22B against another model?
Select another target model to view immediate pricing & latency trade-offs.