The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into French against 10 requirements in three levels: the solutions that meet them, what they cost and how fast they are, from the 2026 benchmark of 57 AI models and MT engines.
A solution is a model plus its harness: prompts, glossaries and other context, pre- and post-processing. It meets a quality level when it meets every requirement of the level.
| Requirement | Description | Met when |
|---|---|---|
| Basic requirements for any company | ||
| General translation quality | Translation errors of every type, such as mistranslations, omissions and grammar errors, classified by type and severity following MQM, the industry standard. | No critical error, at most 3 major errors in the four test texts (3,034 words) and at most 20 errors per 1,000 words, weighted by severity (minor 1, major 5, critical 10). |
| Standard requirements adds the rules companies set for their content | ||
| Terminology | The approved translation of every glossary term, in any of its correct forms. | 100% of the glossary terms. |
| Tone of voice | The formal register (vous) in every sentence in the second person or the imperative. | 100% of those sentences. |
| Tags | Every HTML tag of the source, with its attributes, in the translation. | 100% of the tags. |
| Unicode character integrity | Every character intact, with no corrupted characters and no letter from another script that only looks like the right one. | 100% of the characters. |
| Locale conventions | The punctuation marks of the target language in place of the English ones, such as guillemets. | No pass-or-fail bar. We report it as a share, because whether to localize punctuation is a matter of house style. |
| HTML entities | No HTML character code, such as the code for a quotation mark, that the source does not carry. | No added HTML entity. |
| Valid JSON | The translated JSON document still parses. | The document parses. |
| Spacing at tag boundaries | The spaces next to HTML tags stay as in the source. | 100% of the tag boundaries. |
| Advanced requirements adds a requirement for some companies and use cases | ||
| Gender neutrality | A gender-neutral translation wherever the English source leaves the gender of a person open, as in “your manager” or “the expert”. | 100% of the sentences that refer to a person. |
Pick a quality level, or your own requirements, bars and constraints, to see the shortlisted solutions that meet them in French, with a column for each requirement you select. Under Optimize for, choose which of them comes first: the one with the fewest total errors, the lowest cost or the shortest time. Click a column heading to sort by it.
The shortlisted solutions on 13 models, with their Intento LQA results. Total errors per 1,000 words add the weighted general translation quality errors (minor 1, major 5, critical 10) to the missed glossary terms, sentences in the wrong register, tag errors and sentences that fail gender neutrality, each counted once. Formatting counts as met or missed and does not add to the total errors. The models that did not make the shortlist have automatic scores only: the shortlist. Gender neutrality: the Advanced requirements count the two best solutions of the language, with the gender-neutrality agent where they need it. The results of the other solutions are shown for information.
Two three-step solutions (translation + LQA + repair) meet every Standard requirement in French, and neither leads on errors, cost and time at once. The one on Anthropic Claude Fable 5 has 24.4 total errors per 1,000 words against 26.3 for the one on Google Gemini 3.1 Pro Preview, and takes 173 against 286 seconds per 1,000 words. It costs $604 against $576 per million characters. Optimizing for quality or latency selects Anthropic Claude Fable 5, and optimizing for cost selects Google Gemini 3.1 Pro Preview.
None of the solutions that meet every Basic requirement in French cost less than $575 per million characters. None of the real-time, one-step or two-step solutions (translation + post-edit) meet even the Basic requirements. The Basic requirements cover general translation quality alone, and under them the three-step solution (translation + LQA + repair) on OpenAI GPT-4.1 takes the shortest time, 107 seconds per 1,000 words, at $584 per million characters.
For the Advanced requirements, the gender-neutrality agent works with both three-step solutions (translation + LQA + repair) that meet every Standard requirement. On Anthropic Claude Fable 5 it raises the cost from $604 to $995 per million characters and the time from 173 to 191 seconds per 1,000 words, and total errors fall from 24.4 to 17.2 per 1,000 words. On Google Gemini 3.1 Pro Preview the cost goes from $576 to $964 per million characters and total errors from 26.3 to 20.1 per 1,000 words.
Anthropic Claude Fable 5
three-step
Google Gemini 3.1 Pro Preview
three-step
OpenAI GPT-4.1
three-step
Anthropic Claude Fable 5
three-step
Google Gemini 3.1 Pro Preview
three-step
Anthropic Claude Fable 5
three-step
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Misses gender neutrality.
Optimize for selects one of the solutions that meet the level: the one with the fewest total errors, the lowest cost or the shortest time. Where no solution meets the level, the card shows the available solution with the fewest total errors and what it misses. Total errors count every requirement with countable errors.
| Quality level | Tier | Solutions | Simplest solution | Model | Total errors per 1,000 words | Cost per million characters | Seconds per 1,000 words |
|---|---|---|---|---|---|---|---|
| Basic | Tier 2: two or more solutions, none simple | ☆×4 | three-step | Google Gemini 3.1 Pro Preview | 26.3 | $576 | 286 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | three-step | Google Gemini 3.1 Pro Preview | 26.3 | $576 | 286 |
| Advanced | Tier 4: no solution | No solution | — | — | — |
One solution per level: the simplest that meets the level, the cheapest of them where several are equally simple. ★ a one-step solution, ☆ a solution with more steps or the gender-neutrality agent.
Four three-step solutions (translation + LQA + repair) meet every Basic requirement in French, and the Standard requirements remove two of them on terminology. Those on OpenAI GPT-5.5 and OpenAI GPT-4.1 get 94.4% and 94.0% of the glossary terms right. For both sets of requirements, optimizing for quality selects Anthropic Claude Fable 5 and optimizing for cost Google Gemini 3.1 Pro Preview. Optimizing for latency selects OpenAI GPT-4.1 under the Basic requirements and Anthropic Claude Fable 5 under the Standard ones.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-5.6 Luna | 0.4 | 32.3 | Misses a Standard requirement |
| Mistral Large 3 | 0.73 | 35.3 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.82 | 43.5 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.09 | 31.3 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.16 | 26.7 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 5.72 | 39.2 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 9.89 | 26 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 14.49 | 24.4 | Misses a Standard requirement |
| Amazon Translate | 15 | 79.8 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 44.2 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 28.78 | 22.1 | Misses a Standard requirement |
Among the one-step solutions in French with a cost, the three dearest on LLMs also have the fewest errors: Anthropic Claude Fable 5 at $28.78 per million characters and 22.1 weighted errors per 1,000 words, Anthropic Claude Opus 4.8 at $14.49 and 24.4, OpenAI GPT-5.5 at $9.89 and 26.0. Those on DeepL Next Gen and Amazon Translate cost $27.50 and $15 per million characters, more than every LLM solution on the chart but Anthropic Claude Fable 5. They have 44.2 and 79.8 weighted errors per 1,000 words, more than any LLM solution there.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 33.9 |
| Gender neutrality | 9.1 |
| Terminology | 0.2 |
| Tone of voice | 0 |
| Tag handling | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
The steps in French depend on general translation quality. The one-step solutions closest to its bar, on Anthropic Claude Fable 5, Anthropic Claude Opus 4.8, OpenAI GPT-5.5 and Google Gemini 3.1 Pro Preview, have 22.1 to 26.7 weighted errors per 1,000 words. Two-step solutions (translation + post-edit) on three of these models have 22.1 to 25.7 weighted errors per 1,000 words. Three-step solutions (translation + LQA + repair) bring three of these models and OpenAI GPT-4.1 under the bar, at 15.5 to 19.8 weighted errors per 1,000 words.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in French | Class |
|---|---|---|
| Amazon Translate | 79.8 | MT engines |
| OpenAI gpt-oss-120b | 54.1 | Open-weight LLMs |
| DeepL Next Gen | 44.2 | LLM-based MT engines |
| OpenAI GPT-4.1 | 43.5 | LLMs |
| Anthropic Claude Sonnet 5 | 39.2 | LLMs |
| Mistral Large 3 | 35.3 | LLMs |
| Cohere Command A Translate | 33.9 | LLMs |
| OpenAI GPT-5.6 Luna | 32.3 | LLMs |
| Google Gemini 3.5 Flash | 31.3 | LLMs |
| Google Gemini 3.1 Pro Preview | 26.7 | LLMs |
| OpenAI GPT-5.5 | 26 | LLMs |
| Anthropic Claude Opus 4.8 | 24.4 | LLMs |
| Anthropic Claude Fable 5 | 22.1 | LLMs |
No translation into French: Widn AI Tower Vesuvius (cut off part-way).
Translations from the ten shortlisted LLMs other than the open-weight one have 22.1 to 43.5 weighted errors per 1,000 words in one step in French. Translations from Amazon Translate, the one shortlisted MT engine that is not LLM-based, have 79.8 weighted errors per 1,000 words and 5 critical errors, the only critical errors among the shortlisted one-step solutions. All 11 shortlisted LLMs, the open-weight OpenAI gpt-oss-120b among them, support French by our measured criteria.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Sonnet 4.6 | 0.779 | 11 | 1.34 | 10 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.778 | 12 | 1.36 | 13 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.778 | 13 | 1.37 | 16 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.777 | 14 | 1.39 | 20 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.777 | 15 | 1.36 | 15 | No | Not shortlisted |
| z.ai GLM-5 | 0.776 | 17 | 1.35 | 12 | No | Not shortlisted |
| Google Gemma 4 | 0.776 | 18 | 1.36 | 14 | No | Not shortlisted |
| Supertext LLM | 0.775 | 19 | 1.39 | 21 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.772 | 21 | 1.68 | 36 | No | Not shortlisted |
| Widn AI Tower Sugarloaf | 0.771 | 23 | 1.34 | 11 | No | Not shortlisted |
| Meta Llama 4 Maverick | 0.767 | 24 | 1.47 | 22 | No | Not shortlisted |
| Microsoft Translator | 0.765 | 25 | 1.50 | 24 | No | Not shortlisted |
| DeepSeek V3.2 | 0.765 | 26 | 1.52 | 25 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.765 | 27 | 1.53 | 27 | No | Not shortlisted |
| Google Translation LLM | 0.764 | 28 | 1.84 | 38 | No | Not shortlisted |
| MiniMax M2.5 | 0.760 | 29 | 1.72 | 37 | No | Not shortlisted |
| Translated Lara 2 | 0.758 | 30 | 1.64 | 34 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.758 | 31 | 1.57 | 31 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.756 | 32 | 1.55 | 30 | No | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | 0.755 | 33 | 1.58 | 32 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.755 | 34 | 1.53 | 28 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.754 | 35 | 1.61 | 33 | No | Not shortlisted |
| Elia Machine Translation System | 0.752 | 36 | 1.53 | 26 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.749 | 37 | 1.54 | 29 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.749 | 38 | 1.68 | 35 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.740 | 39 | 2.29 | 44 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.731 | 40 | 1.86 | 39 | No | Not shortlisted |
| Cohere Tiny Aya | 0.731 | 41 | 2.50 | 45 | No | Not shortlisted |
| iFlyTek MT | 0.723 | 42 | 2.16 | 42 | No | Not shortlisted |
| Tencent Cloud TMT | 0.721 | 43 | 2.09 | 41 | No | Not shortlisted |
| Naver Papago NMT Commercial | 0.720 | 44 | 2.05 | 40 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.711 | 45 | 2.24 | 43 | No | Not shortlisted |
| AppTek NMT | 0.664 | 46 | 3.43 | 46 | No | Not shortlisted |
| Widn AI Tower Anthill | 0.434 | 47 | 10.88 | 47 | No | Not shortlisted |
| DeepL Next Gen | 0.787 | 1 | 1.21 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.784 | 2 | 1.37 | 17 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.5 | 0.783 | 3 | 1.31 | 5 | Yes | Top 8 by both metrics |
| Cohere Command A Translate | 0.782 | 4 | 1.32 | 7 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.782 | 5 | 1.30 | 4 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.781 | 6 | 1.39 | 19 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | 0.780 | 7 | 1.33 | 9 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Fable 5 | 0.780 | 8 | 1.33 | 8 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.779 | 9 | 1.28 | 3 | Yes | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | 0.779 | 10 | 1.32 | 6 | Yes | Top 8 by MetricX |
| Amazon Translate | 0.776 | 16 | 1.37 | 18 | Yes | Real-time place |
| Mistral Large 3 | 0.772 | 20 | 1.26 | 2 | Yes | Top 8 by MetricX |
| OpenAI gpt-oss-120b | 0.771 | 22 | 1.49 | 23 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Adjusted NMT | 1 | 0.787 | 1 | 1.21 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 2 | 0.784 | 17 | 1.37 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.5 | Adjusted LLM | 3 | 0.783 | 5 | 1.31 | Top 8 by both metrics |
| Cohere Command A Translate | Adjusted LLM | 4 | 0.782 | 7 | 1.32 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 5 | 0.782 | 4 | 1.30 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 6 | 0.781 | 19 | 1.39 | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 7 | 0.780 | 9 | 1.33 | Top 8 by COMET-Kiwi |
| Anthropic Claude Fable 5 | Adjusted LLM | 8 | 0.780 | 8 | 1.33 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 9 | 0.779 | 3 | 1.28 | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 10 | 0.779 | 6 | 1.32 | Top 8 by MetricX |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 11 | 0.779 | 10 | 1.34 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 12 | 0.778 | 13 | 1.36 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 13 | 0.778 | 16 | 1.37 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.777 | 20 | 1.39 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 15 | 0.777 | 15 | 1.36 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 16 | 0.776 | 18 | 1.37 | Real-time place |
| z.ai GLM-5 | Adjusted LLM | 17 | 0.776 | 12 | 1.35 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 18 | 0.776 | 14 | 1.36 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 19 | 0.775 | 21 | 1.39 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 20 | 0.772 | 2 | 1.26 | Top 8 by MetricX |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 21 | 0.772 | 36 | 1.68 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 22 | 0.771 | 23 | 1.49 | Real-time place |
| Widn AI Tower Sugarloaf | Adjusted NMT | 23 | 0.771 | 11 | 1.34 | Not shortlisted |
| Meta Llama 4 Maverick | Adjusted LLM | 24 | 0.767 | 22 | 1.47 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 25 | 0.765 | 24 | 1.50 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 26 | 0.765 | 25 | 1.52 | Not shortlisted |
| Tilde TildeOpen LLM engine | Adjusted NMT | 27 | 0.765 | 27 | 1.53 | Not shortlisted |
| Google Translation LLM | Off-the-shelf NMT | 28 | 0.764 | 38 | 1.84 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 29 | 0.760 | 37 | 1.72 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 30 | 0.758 | 34 | 1.64 | Not shortlisted |
| SYSTRAN PNMT API | Adjusted NMT | 31 | 0.758 | 31 | 1.57 | Not shortlisted |
| Yandex Cloud Translate v2 | Adjusted NMT | 32 | 0.756 | 30 | 1.55 | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | Adjusted LLM | 33 | 0.755 | 32 | 1.58 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 34 | 0.755 | 28 | 1.53 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 35 | 0.754 | 33 | 1.61 | Not shortlisted |
| Elia Machine Translation System | Off-the-shelf NMT | 36 | 0.752 | 26 | 1.53 | Not shortlisted |
| ModernMT Enterprise Edition | Adjusted NMT | 37 | 0.749 | 29 | 1.54 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 38 | 0.749 | 35 | 1.68 | Not shortlisted |
| Google Cloud Advanced Translation API | Adjusted NMT | 39 | 0.740 | 44 | 2.29 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 40 | 0.731 | 39 | 1.86 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 41 | 0.731 | 45 | 2.50 | Not shortlisted |
| iFlyTek MT | Off-the-shelf NMT | 42 | 0.723 | 42 | 2.16 | Not shortlisted |
| Tencent Cloud TMT | Off-the-shelf NMT | 43 | 0.721 | 41 | 2.09 | Not shortlisted |
| Naver Papago NMT Commercial | Off-the-shelf NMT | 44 | 0.720 | 40 | 2.05 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 45 | 0.711 | 43 | 2.24 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 46 | 0.664 | 46 | 3.43 | Not shortlisted |
| Widn AI Tower Anthill | Adjusted NMT | 47 | 0.434 | 47 | 10.88 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.774 | — | 1.36 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.780 | — | 1.29 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.772 | — | 1.45 | Same model family as xAI Grok 4.20 (reasoning) |
Every model we got a translation from, as the shortlist compared it: off the shelf, without context in the request, or adjusted, with it. A model that did not make the shortlist is shown adjusted, unless off the shelf it scores higher by 0.03 or more on COMET-Kiwi. Ranks among 47 models, with each model family counted once, and the top 8 of each metric in bold. Higher is better on COMET-Kiwi, lower on MetricX. The scores rank models within this language and are not a requirement result.
MetricX ranks Mistral Large 3 second in French and COMET-Kiwi ranks it 20th, while COMET-Kiwi ranks OpenAI GPT-5.6 Luna second and MetricX 17th, so each is on the shortlist through one metric. By Intento LQA, the one-step solutions on them have 35.3 and 32.3 weighted errors per 1,000 words, against 26.7 for Google Gemini 3.1 Pro Preview, seventh on COMET-Kiwi and ninth on MetricX. Nearest the cut in the tail is Anthropic Claude Sonnet 4.6, 11th on COMET-Kiwi and 10th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, three-step | $576 |
| Standard | Google Gemini 3.1 Pro Preview, three-step | $576 |
| Advanced | No solution | — |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.4 | LLMs |
| Mistral Large 3 | $0.73 | LLMs |
| OpenAI GPT-4.1 | $2.82 | LLMs |
| Google Gemini 3.5 Flash | $3.09 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.16 | LLMs |
| Anthropic Claude Sonnet 5 | $5.72 | LLMs |
| OpenAI GPT-5.5 | $9.89 | LLMs |
| Anthropic Claude Opus 4.8 | $14.49 | LLMs |
| Amazon Translate | $15 | MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $28.78 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.404 | LLMs |
| Anthropic Claude Sonnet 5 | 0.404 | LLMs |
| Anthropic Claude Fable 5 | 0.402 | LLMs |
| Mistral Medium 3.1 | 0.322 | LLMs |
| Mistral Large 3 | 0.315 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.305 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.304 | LLMs |
| MoonshotAI Kimi K2.5 | 0.285 | LLMs |
| AI21 Jamba Large 1.7 | 0.277 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.274 | Open-weight LLMs |
| Widn AI Tower Sugarloaf | 0.272 | LLM-based MT engines |
| Widn AI Tower Anthill | 0.269 | LLM-based MT engines |
| DeepSeek V3.2 | 0.267 | LLMs |
| z.ai GLM-5 | 0.258 | LLMs |
| Cohere Command A Translate | 0.251 | LLMs |
| Google Gemma 4 | 0.245 | Open-weight LLMs |
| Google Gemini 3.1 Pro Preview | 0.244 | LLMs |
| Google Gemini 3.5 Flash | 0.244 | LLMs |
| OpenAI GPT-4.1 | 0.239 | LLMs |
| Cohere Tiny Aya | 0.238 | LLMs |
| OpenAI GPT-5.6 Luna | 0.236 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.235 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.235 | LLMs |
| OpenAI GPT-5.5 | 0.235 | LLMs |
| OpenAI GPT-5.6 Terra | 0.235 | LLMs |
| OpenAI GPT-5.6 Sol | 0.234 | LLMs |
| MiniMax M2.5 | 0.22 | LLMs |
The tokenizers of the three Anthropic Claude models on the shortlist produce 0.402 to 0.404 tokens per character of the French translation, against 0.235 for OpenAI GPT-5.5. Even so, the one-step solution on Anthropic Claude Sonnet 5 costs $5.72 per million characters, less than the $9.89 of the one on OpenAI GPT-5.5. On one model, the cost follows the steps: the one-step solution on Google Gemini 3.1 Pro Preview costs $4.16 per million characters, and the three-step one (translation + LQA + repair) $576.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-4.1, three-step | 107 |
| Standard | Anthropic Claude Fable 5, three-step | 173 |
| Advanced | No solution | — |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| Google Translation LLM | 0.4 | LLM-based MT engines |
| Pangeanic PangeaMT | 0.5 | MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| ModernMT Enterprise Edition | 1.1 | MT engines |
| Amazon Translate | 1.3 | MT engines |
| Translated Lara 2 | 1.3 | LLM-based MT engines |
| SYSTRAN PNMT API | 1.5 | MT engines |
| OpenAI gpt-oss-120b | 1.7 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3 | MT engines |
| Oracle Cloud Language Service | 4.5 | MT engines |
| Elia Machine Translation System | 6.8 | MT engines |
| Widn AI Tower Anthill | 7.7 | LLM-based MT engines |
| Tilde TildeOpen LLM engine | 9.3 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 12 | LLMs |
| Google Gemini 3.5 Flash | 13 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 13 | LLMs |
| OpenAI GPT-4.1 | 15 | LLMs |
| Supertext LLM | 16 | LLM-based MT engines |
| Alibaba Machine Translation API | 18 | MT engines |
| Cohere Tiny Aya | 19 | LLMs |
| OpenAI GPT-5.6 Terra | 20 | LLMs |
| Anthropic Claude Haiku 4.5 | 25 | LLMs |
| Cohere Command A Translate | 29 | LLMs |
| Mistral Medium 3.1 | 31 | LLMs |
| Swiss AI Apertus-8B-2509 | 35 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 36 | LLMs |
| Mistral Large 3 | 36 | LLMs |
| DeepSeek V3.2 | 38 | LLMs |
| MiniMax M2.5 | 40 | LLMs |
| AI21 Jamba Large 1.7 | 43 | LLMs |
| xAI Grok 4.20 (reasoning) | 44 | LLMs |
| OpenAI GPT-5.6 Sol | 45 | LLMs |
| Anthropic Claude Opus 4.8 | 46 | LLMs |
| Anthropic Claude Sonnet 4.6 | 46 | LLMs |
| OpenAI GPT-5.5 | 47 | LLMs |
| Meta Llama 4 Maverick | 48 | Open-weight LLMs |
| Anthropic Claude Fable 5 | 49 | LLMs |
| iFlyTek MT | 54 | MT engines |
| Tencent Cloud TMT | 55 | MT engines |
| SpeakLeash Bielik-11B-v3.0-Instruct | 56 | Open-weight LLMs |
| Widn AI Tower Sugarloaf | 61 | LLM-based MT engines |
| AppTek NMT | 90 | MT engines |
| Google Gemini 3.1 Pro Preview | 94 | LLMs |
| Alibaba Qwen3-235B-A22B | 100 | Open-weight LLMs |
| Naver Papago NMT Commercial | 138 | MT engines |
| z.ai GLM-5 | 166 | LLMs |
| Google Gemma 4 | 167 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 190 | LLMs |
Real-time models are those the shortlist counts as real-time, under 2 seconds per text across the languages we timed. The chart shows the time of each model into French.
The real-time models in French are six MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. Three of them are on the shortlist, DeepL Next Gen, Amazon Translate and OpenAI gpt-oss-120b, and translations from all three meet tone of voice. None of the three meet general translation quality, terminology or tag handling. The three-step solutions (translation + LQA + repair) that meet every Basic requirement take 107 to 458 seconds per 1,000 words.
| The two best solutions | Solution | Sentences about people that are gender-neutral | Meets every Advanced requirement | Cost per million characters | Seconds per 1,000 words |
|---|---|---|---|---|---|
| Anthropic Claude Fable 5 | three-step, with the gender-neutrality agent | 98.2% | ✗ | $995 | 191 |
| Google Gemini 3.1 Pro Preview | three-step, with the gender-neutrality agent | 99.1% | ✗ | $964 | 285 |
For French we check the formal register, with vous, and gender-neutral wording through neutral nouns and rephrasing, with doubling only when unavoidable, no glyphs and no default gender. With the gender-neutrality agent, the three-step solutions (translation + LQA + repair) on Anthropic Claude Fable 5 and Google Gemini 3.1 Pro Preview meet every other Advanced requirement and miss only gender neutrality, where we require 100% of the sentences about people to be gender-neutral.
For the 13 shortlisted solutions in French, the median share of sentences on which the three LQA judges agree is 97.7%. With agreement that high, the Intento LQA error counts on this page are reliable.
Leave feedback