Best AI models for French
French text uses more tokens than English, and the gap depends on the model. We measured it ourselves, then combined it with independent quality scores and current prices to show which models give the best results for your money in French.
The cost of French, by model family
| Tokenizeri | Tokens (English) | Tokens (French) | Extra vs English |
|---|---|---|---|
| OpenAI (GPT-4o) | 219 | 273 | +25% |
| Mistral (Nemo) | 220 | 280 | +27% |
| Meta (Llama 4) | 219 | 279 | +27% |
| Google (Gemma 3) | 218 | 287 | +32% |
| Z.ai (GLM-4.5) | 219 | 304 | +39% |
| Alibaba (Qwen 3) | 219 | 325 | +48% |
| DeepSeek (V3.1) | 219 | 327 | +49% |
Anthropic (Claude) and xAI (Grok) do not publish their tokenizers, so they cannot be measured; for their models we use the average of the measured families (marked ≈). Google’s Gemini is estimated with Google’s open Gemma tokenizer.
Most capable models and their real cost in French
Rankings →| Model | Overall indexi | French factor | Inputi | 1,000 pages |
|---|---|---|---|---|
GPT-6 AstraOpenAI | 166.6 | ×1.25 | $10 | $4.22 |
Claude Fable 5.1Anthropic | 165.0 | ≈ ×1.35 | $10 | $4.58 |
Claude Fable 5Anthropic | 163.6 | ≈ ×1.35 | $10 | $4.58 |
Claude Opus 5Anthropic | 162.7 | ≈ ×1.35 | $5 | $2.29 |
GPT-5.5 ProOpenAI | 162.5 | ×1.25 | $30 | $12.66 |
GPT-5.6 SolOpenAI | 162.0 | ×1.25 | $2 | $0.844 |
GPT-5.6 TerraOpenAI | 159.3 | ×1.25 | $2 | $0.844 |
GPT-5.5OpenAI | 159.3 | ×1.25 | $5 | $2.11 |
GPT-5.4 ProOpenAI | 159.1 | ×1.25 | $30 | $12.66 |
Claude Opus 4.8Anthropic | 158.3 | ≈ ×1.35 | $5 | $2.29 |
Gemini 3.7 FlashGoogle DeepMind | 157.7 | ×1.32 | $0.75 | $0.333 |
Kimi K3Moonshot AI | 157.7 | ≈ ×1.35 | $3 | $1.38 |
Gemini 3.8 FlashGoogle DeepMind | 157.1 | ×1.32 | $0.75 | $0.333 |
Muse Spark 1.3Meta | 156.9 | ×1.27 | $1.25 | $0.539 |
GPT-5.4OpenAI | 156.9 | ×1.25 | $2.50 | $1.06 |
Cost to read 1,000 pages of text (about 300,000 words in English) written in French, as input. Each model gets the factor measured on its company’s latest public tokenizer, so treat it as an estimate: newer models may use a different tokenizer. ≈ marks companies that publish no tokenizer (average of measured families).
Best value for French
Strong models (overall index 140 or more), cheapest first for processing French text.
- $0.0091 / 1,000 pages
- $0.015 / 1,000 pages
- $0.033 / 1,000 pages
- $0.034 / 1,000 pages
- $0.04 / 1,000 pages
- $0.068 / 1,000 pages
- $0.071 / 1,000 pages
- $0.075 / 1,000 pages
Models built for French
Specialised models are not always available through the big APIs above, but they are worth testing, especially for local dialects and cultural knowledge.
- Mistral AI ↗Paris-based lab with open-weight and commercial models, strong in French and other European languages.
- Lucie ↗Open-source model trained with a large share of French data.
- CroissantLLM ↗Small bilingual French–English model trained on equal amounts of both languages.
- Pleias ↗French startup releasing small open models trained on openly licensed data.
Independent French leaderboards
These projects test models directly in French. Use them alongside the overall scores above.
- compar:IA ↗Blind "which answer is better" comparisons in French, with a public ranking and energy estimates.
- Leaderboard LLM FR ↗Benchmarks of language models on French-language tests.
Tips for using AI in French
- Set the tone. Say whether the model should use "tu" or "vous"; many models default to a formal register, which can sound stiff in casual content.
- Name your audience. French in France, Quebec, Belgium, Switzerland or Africa differs in vocabulary and expressions: tell the model who you are writing for.
- Watch for anglicisms. Translated or generated French can sound unnatural ("faire du sens"): ask for idiomatic French and proofread.
- Budget for extra tokens. Depending on the model, the same content costs 25% to 49% more in French than in English: use the table above or the token counter to estimate.
- Check where your data goes. For personal or sensitive data, check where the provider processes it and whether that fits your GDPR obligations.
How we measured
We wrote the same four texts (a news item, a customer email, a technical explanation and travel tips) in 15 languages, then counted the tokens that each public tokenizer produces. Quality scores come from Epoch AI and are measured mostly in English, so always test the finalists on your own French content.
Get a personal recommendationCount the tokens of your own text