Side-by-side technical and economic comparison between Mistral AI's Codestral (Current Generation) and Meta / Together's Llama 3.1 405B — Together AI.
• Llama 3.1 405B — Together AI: Classified as a Legacy Reference model (Deprecated on Together AI (Reference)).
| Metric | Codestral (Current Generation) | Llama 3.1 405B — Together AI | Advantage |
|---|---|---|---|
| Provider Organization | Mistral AI | Meta / Together | — |
| Standard Input / 1M | $0.30 | $3.50 | Codestral (Current Generation) (91% lower) |
| Cached Input / 1M | $0.03 | None | Codestral (Current Generation) lower |
| Output / 1M | $0.90 | $3.50 | Codestral (Current Generation) lower |
| Context Window | 256k | 128k | Codestral (Current Generation) (256k) |
| Max Generation Tokens | 8.2k | 4.1k | Codestral (Current Generation) |
| Tokenizer Family | Mistral Tekken BPE (~131k vocabulary) | Meta Llama 3 Tiktoken BPE (~128k vocabulary) | — |
Opt for Codestral (Current Generation) when your engineering requirements prioritize Mistral AI's ecosystem, specific tokenizer efficiencies (Exact Tekken Tokenizer (±2%)), or when your expected prompt-to-completion ratios favor its $0.3/M input rate.
Opt for Llama 3.1 405B — Together AI when looking for Meta / Together's tooling integration, specific context window depth (128k tokens), or when output generation volume favors its $3.5/M rate.