Median one-step costs are $4.87 per million source characters for LLMs, $10 for MT engines and $25 for LLM-based MT engines.
Among LLMs, OpenAI GPT-5.6 Luna costs the least, $0.46 per million characters at the median, and among LLM-based MT engines, the cost runs from $10 on Widn AI Tower Sugarloaf to $27.50 on DeepL Next Gen.
Cost per million characters of the one-step solutions, by model
| Model | Type | Median cost per million characters | Languages on the shortlist |
|---|---|---|---|
| ALL-Lab | MT engine | $100.00 | 3 |
| Anthropic Claude Fable 5 | LLM | $32.84 | 72 |
| DeepL Next Gen | LLM-based MT engine | $27.50 | 81 |
| Translated Lara 2 | LLM-based MT engine | $24.99 | 71 |
| Supertext LLM | LLM-based MT engine | $23.40 | 1 |
| Google Translation LLM | LLM-based MT engine | $21.35 | 38 |
| Google Cloud Advanced Translation API | MT engine | $20.00 | 7 |
| Anthropic Claude Opus 4.8 | LLM | $16.07 | 55 |
| Amazon | MT engine | $15.00 | 7 |
| OpenAI GPT-5.5 | LLM | $11.64 | 71 |
| Microsoft Translator | MT engine | $10.00 | 40 |
| Widn AI Tower Sugarloaf | LLM-based MT engine | $10.00 | 11 |
| OpenAI GPT-5.6 Sol | LLM | $7.92 | 35 |
| Anthropic Claude Sonnet 4.6 | LLM | $7.37 | 43 |
| Anthropic Claude Sonnet 5 | LLM | $6.09 | 42 |
| Google Gemini 3.1 Pro Preview | LLM | $4.97 | 88 |
| OpenAI GPT-5.6 Terra | LLM | $4.66 | 26 |
| Yandex | MT engine | $3.94 | 2 |
| Google Gemini 3.5 Flash | LLM | $3.74 | 77 |
| AI21 Jamba Large 1.7 | LLM | $3.62 | 3 |
| OpenAI GPT-4.1 | LLM | $3.11 | 59 |
| z.ai GLM-5 | LLM | $2.32 | 44 |
| Anthropic Claude Haiku 4.5 | LLM | $2.15 | 5 |
| xAI Grok 4.20 (non-reasoning) | LLM | $1.29 | 16 |
| MoonshotAI Kimi K2.5 | LLM | $1.27 | 13 |
| xAI Grok 4.20 (reasoning) | LLM | $1.26 | 21 |
| DeepSeek V3.2 | LLM | $1.08 | 23 |
| Alibaba Cloud Qwen3-235B-A22B | open-weight LLM | $1.08 | 2 |
| Mistral Medium 3.1 | LLM | $1.02 | 6 |
| Mistral Large 3 | LLM | $0.83 | 30 |
| MiniMax M2.5 | LLM | $0.50 | 1 |
| OpenAI GPT-5.6 Luna | LLM | $0.46 | 33 |
| Google Gemma 4 | open-weight LLM | $0.14 | 45 |
Median over the languages where the model is on the shortlist, translated without context on MT engines off the shelf and with context otherwise. For models with reasoning, the cost is a lower bound.