The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Hindi 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. We test it in the 16 languages with a glossary, so it does not apply in Hindi. |
| Tone of voice | The formal register (आप) 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. | 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 Hindi, 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.
Three two-step solutions (translation + post-edit) meet every Standard requirement in Hindi, and none leads on errors, cost and time at once. The one on Anthropic Claude Fable 5 has 23.7 total errors per 1,000 words and takes 66 seconds per 1,000 words, at $37.80 per million characters. The one on OpenAI GPT-5.6 Sol has 24.9 total errors per 1,000 words and takes 131 seconds per 1,000 words, at $7.45 per million characters. Optimizing for quality or latency selects Anthropic Claude Fable 5, and for cost OpenAI GPT-5.6 Sol.
Only the two-step solution on OpenAI GPT-5.6 Sol meets every Standard requirement in Hindi for less than $27.85 per million characters. Only the one on Anthropic Claude Fable 5 meets them in less than 131 seconds per 1,000 words. With a limit of one step, the solutions on these two models meet every Basic requirement, but not every Standard requirement. The third, the two-step solution on Google Gemini 3.1 Pro Preview, costs $27.86 per million characters and takes the longest, 178 seconds per 1,000 words.
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
two-step
OpenAI GPT-5.6 Sol
two-step
Anthropic Claude Fable 5
two-step
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
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 1: simple solution + backup | ★★ ☆ | one-step | OpenAI GPT-5.6 Sol | 25.3 | $7.45 | 131 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆☆ | two-step | OpenAI GPT-5.6 Sol | 24.9 | $7.45 | 131 |
| Advanced | Tier 3: one solution | ☆ | with the gender-neutrality agent | Anthropic Claude Fable 5 | 6.9 | $1,045 | 228 |
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.
The Standard requirements keep the three models that meet every Basic requirement in Hindi, all as two-step solutions (translation + post-edit). Those on Anthropic Claude Fable 5 and OpenAI GPT-5.6 Sol meet every Basic requirement in one step, and tag handling with the post-edit step. Optimizing for cost selects OpenAI GPT-5.6 Sol under both, and for quality or latency Anthropic Claude Fable 5, in one step under the Basic requirements and in two steps under the Standard ones.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.12 | 26.4 | Misses a Standard requirement |
| Mistral Large 3 | 1.51 | 41.9 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.02 | 58.3 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.11 | 34.3 | Misses a Standard requirement |
| z.ai GLM-5 | 3.81 | 32 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.09 | 23.1 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 7.45 | 12.2 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 7.6 | 30 | Misses a Standard requirement |
| Microsoft Translator | 10 | 103.5 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 19 | 26.7 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 27.4 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 39.9 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 37.8 | 11.5 | Misses a Standard requirement |
The one-step solutions in Hindi on Translated Lara 2 and DeepL Next Gen, both LLM-based MT engines, cost $24.99 and $27.50 per million characters, with 27.4 and 39.9 weighted errors per 1,000 words. The one on the open-weight Google Gemma 4 costs $0.12 per million characters and has 26.4 weighted errors per 1,000 words. Of the solutions that meet every Standard requirement, the dearest, on Anthropic Claude Fable 5, has 11.5 weighted errors per 1,000 words, against 12.2 for the one on Google Gemini 3.1 Pro Preview.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 30 |
| Gender neutrality | 12.9 |
| Tag handling | 1.3 |
| Tone of voice | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Of the Standard requirements, general translation quality and tag handling are the bottleneck for one-step solutions in Hindi. The ones on Google Gemini 3.1 Pro Preview and z.ai GLM-5 get every tag right, with 23.1 and 32.0 weighted errors per 1,000 words and 4 and 5 major errors. Two-step solutions (translation + post-edit) meet both requirements on three models. On Google Gemini 3.1 Pro Preview the weighted errors fall to 12.2 per 1,000 words, and on Anthropic Claude Fable 5 and OpenAI GPT-5.6 Sol every tag comes out right.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Hindi | Class |
|---|---|---|
| Microsoft Translator | 103.5 | MT engines |
| OpenAI GPT-4.1 | 58.3 | LLMs |
| Mistral Large 3 | 41.9 | LLMs |
| DeepL Next Gen | 39.9 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 34.3 | LLMs |
| z.ai GLM-5 | 32 | LLMs |
| Anthropic Claude Sonnet 5 | 30 | LLMs |
| Translated Lara 2 | 27.4 | LLM-based MT engines |
| Anthropic Claude Opus 4.8 | 26.7 | LLMs |
| Google Gemma 4 | 26.4 | Open-weight LLMs |
| Google Gemini 3.1 Pro Preview | 23.1 | LLMs |
| OpenAI GPT-5.6 Sol | 12.2 | LLMs |
| Anthropic Claude Fable 5 | 11.5 | LLMs |
No translation into Hindi: Widn AI Tower Anthill (cut off part-way, wrong language, wrong script), Widn AI Tower Vesuvius (cut off part-way).
Translations into Hindi from one-step solutions on the nine shortlisted LLMs that are not open-weight have 11.5 to 58.3 weighted errors per 1,000 words. Those from Translated Lara 2 and DeepL Next Gen, both LLM-based MT engines, have 27.4 and 39.9 weighted errors per 1,000 words. Those from Microsoft Translator, an MT engine that is not LLM-based, have 103.5 weighted errors per 1,000 words and 40 major errors. The shortlisted LLM that does not meet our language-support criteria is OpenAI GPT-4.1, with 3 critical errors in one step.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.5 | 0.702 | 9 | 1.86 | 13 | No | Not shortlisted |
| DeepSeek V3.2 | 0.701 | 11 | 1.77 | 9 | No | Not shortlisted |
| Cohere Command A Translate | 0.700 | 13 | 1.78 | 10 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.697 | 14 | 1.91 | 19 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.696 | 15 | 1.86 | 14 | No | Not shortlisted |
| MiniMax M2.5 | 0.695 | 16 | 1.87 | 16 | No | Not shortlisted |
| Widn AI Tower Sugarloaf | 0.694 | 17 | 1.87 | 17 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.693 | 18 | 1.93 | 21 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.688 | 21 | 1.85 | 12 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.685 | 23 | 2.04 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.684 | 24 | 2.04 | 24 | No | Not shortlisted |
| Meta Llama 4 Maverick | 0.683 | 25 | 2.05 | 26 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.683 | 26 | 1.87 | 15 | No | Not shortlisted |
| Cohere Tiny Aya | 0.679 | 27 | 1.94 | 22 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.677 | 28 | 2.61 | 33 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.676 | 29 | 2.23 | 29 | No | Not shortlisted |
| Google Translation LLM | 0.673 | 30 | 2.26 | 30 | No | Not shortlisted |
| Amazon Translate | 0.668 | 31 | 2.41 | 31 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.663 | 32 | 2.07 | 27 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.663 | 33 | 2.73 | 34 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.660 | 34 | 2.91 | 35 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.655 | 35 | 2.56 | 32 | No | Not shortlisted |
| Tencent Cloud TMT | 0.645 | 36 | 3.12 | 36 | No | Not shortlisted |
| AppTek NMT | 0.576 | 37 | 4.38 | 38 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.571 | 38 | 4.25 | 37 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.431 | 39 | 9.47 | 39 | No | Not shortlisted |
| DeepL Next Gen | 0.707 | 1 | 1.73 | 6 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | 0.705 | 2 | 1.72 | 5 | Yes | Top 8 by both metrics |
| Google Gemma 4 | 0.705 | 3 | 1.74 | 8 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.705 | 4 | 1.91 | 18 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Fable 5 | 0.705 | 5 | 1.78 | 11 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | 0.704 | 6 | 1.72 | 4 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.703 | 7 | 1.70 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.703 | 8 | 1.92 | 20 | Yes | Top 8 by COMET-Kiwi |
| z.ai GLM-5 | 0.702 | 10 | 1.67 | 2 | Yes | Top 8 by MetricX |
| Mistral Large 3 | 0.700 | 12 | 1.65 | 1 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.690 | 19 | 1.94 | 23 | Yes | Real-time place |
| Microsoft Translator | 0.688 | 20 | 2.17 | 28 | Yes | Real-time place |
| OpenAI GPT-4.1 | 0.687 | 22 | 1.74 | 7 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Off-the-shelf NMT | 1 | 0.707 | 6 | 1.73 | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 2 | 0.705 | 5 | 1.72 | Top 8 by both metrics |
| Google Gemma 4 | Adjusted LLM | 3 | 0.705 | 8 | 1.74 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 4 | 0.705 | 18 | 1.91 | Top 8 by COMET-Kiwi |
| Anthropic Claude Fable 5 | Adjusted LLM | 5 | 0.705 | 11 | 1.78 | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 6 | 0.704 | 4 | 1.72 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 7 | 0.703 | 3 | 1.70 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 8 | 0.703 | 20 | 1.92 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.5 | Adjusted LLM | 9 | 0.702 | 13 | 1.86 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 10 | 0.702 | 2 | 1.67 | Top 8 by MetricX |
| DeepSeek V3.2 | Adjusted LLM | 11 | 0.701 | 9 | 1.77 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 12 | 0.700 | 1 | 1.65 | Top 8 by MetricX |
| Cohere Command A Translate | Adjusted LLM | 13 | 0.700 | 10 | 1.78 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 14 | 0.697 | 19 | 1.91 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 15 | 0.696 | 14 | 1.86 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 16 | 0.695 | 16 | 1.87 | Not shortlisted |
| Widn AI Tower Sugarloaf | Adjusted NMT | 17 | 0.694 | 17 | 1.87 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 18 | 0.693 | 21 | 1.93 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 19 | 0.690 | 23 | 1.94 | Real-time place |
| Microsoft Translator | Off-the-shelf NMT | 20 | 0.688 | 28 | 2.17 | Real-time place |
| Mistral Medium 3.1 | Adjusted LLM | 21 | 0.688 | 12 | 1.85 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 22 | 0.687 | 7 | 1.74 | Top 8 by MetricX |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.685 | 25 | 2.04 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 24 | 0.684 | 24 | 2.04 | Not shortlisted |
| Meta Llama 4 Maverick | Adjusted LLM | 25 | 0.683 | 26 | 2.05 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 26 | 0.683 | 15 | 1.87 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 27 | 0.679 | 22 | 1.94 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 28 | 0.677 | 33 | 2.61 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 29 | 0.676 | 29 | 2.23 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 30 | 0.673 | 30 | 2.26 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 31 | 0.668 | 31 | 2.41 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 32 | 0.663 | 27 | 2.07 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 33 | 0.663 | 34 | 2.73 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 34 | 0.660 | 35 | 2.91 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 35 | 0.655 | 32 | 2.56 | Not shortlisted |
| Tencent Cloud TMT | Off-the-shelf NMT | 36 | 0.645 | 36 | 3.12 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 37 | 0.576 | 38 | 4.38 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 38 | 0.571 | 37 | 4.25 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 39 | 0.431 | 39 | 9.47 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.700 | — | 1.81 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.703 | — | 1.76 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.689 | — | 1.99 | 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 39 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.
The automatic metrics and Intento LQA order the shortlist differently in Hindi. The one-step solutions on the two models the metrics rank first, DeepL Next Gen and Mistral Large 3, have 39.9 and 41.9 weighted errors per 1,000 words. The one-step solution with the fewest, 11.5 weighted errors per 1,000 words, is on Anthropic Claude Fable 5, fifth on COMET-Kiwi and 11th on MetricX. It is on the list through COMET-Kiwi alone. Nearest the cut in the tail are OpenAI GPT-5.5, ninth on COMET-Kiwi, and DeepSeek V3.2, ninth on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | $7.45 |
| Standard | OpenAI GPT-5.6 Sol, two-step | $7.45 |
| Advanced | Anthropic Claude Fable 5, three-step with the gender-neutrality agent | $1,045 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.12 | Open-weight LLMs |
| Mistral Large 3 | $1.51 | LLMs |
| OpenAI GPT-4.1 | $3.02 | LLMs |
| Google Gemini 3.5 Flash | $3.11 | LLMs |
| z.ai GLM-5 | $3.81 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.09 | LLMs |
| OpenAI GPT-5.6 Sol | $7.45 | LLMs |
| Anthropic Claude Sonnet 5 | $7.6 | LLMs |
| Microsoft Translator | $10 | MT engines |
| Anthropic Claude Opus 4.8 | $19 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $37.8 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Mistral Large 3 | 0.889 | LLMs |
| Mistral Medium 3.1 | 0.888 | LLMs |
| AI21 Jamba Large 1.7 | 0.828 | LLMs |
| Widn AI Tower Sugarloaf | 0.786 | LLM-based MT engines |
| Alibaba Qwen3-235B-A22B | 0.784 | Open-weight LLMs |
| z.ai GLM-5 | 0.76 | LLMs |
| Anthropic Claude Sonnet 5 | 0.643 | LLMs |
| Anthropic Claude Fable 5 | 0.64 | LLMs |
| Anthropic Claude Opus 4.8 | 0.637 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.612 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.608 | LLMs |
| Cohere Command A Translate | 0.597 | LLMs |
| DeepSeek V3.2 | 0.516 | LLMs |
| MoonshotAI Kimi K2.5 | 0.503 | LLMs |
| MiniMax M2.5 | 0.465 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.362 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.357 | LLMs |
| OpenAI GPT-4.1 | 0.312 | LLMs |
| OpenAI GPT-5.6 Sol | 0.305 | LLMs |
| OpenAI GPT-5.5 | 0.304 | LLMs |
| OpenAI GPT-5.6 Terra | 0.303 | LLMs |
| OpenAI GPT-5.6 Luna | 0.301 | LLMs |
| Cohere Tiny Aya | 0.288 | LLMs |
| Google Gemini 3.5 Flash | 0.285 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.282 | LLMs |
| Google Gemma 4 | 0.281 | Open-weight LLMs |
Tokens per character alone do not set the one-step costs in Hindi. The tokenizer of Mistral Large 3 produces the most, yet the one-step solution on it costs $1.51 per million characters. The one-step solutions on the shortlisted Anthropic Claude models, at 0.64 tokens per character, cost $7.60 to $37.80 per million characters. On OpenAI GPT-5.6 Sol, the one-step and two-step (translation + post-edit) solutions cost $7.45 per million characters, the three-step one (translation + LQA + repair) with the gender-neutrality agent $983.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Anthropic Claude Fable 5, one-step LLM adjusted for Standard requirements | 66 |
| Standard | Anthropic Claude Fable 5, two-step | 66 |
| Advanced | Anthropic Claude Fable 5, three-step with the gender-neutrality agent | 228 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| Microsoft Translator | 0.3 | MT engines |
| ModernMT Enterprise Edition | 0.5 | MT engines |
| Pangeanic PangeaMT | 0.5 | MT engines |
| DeepL Next Gen | 0.9 | LLM-based MT engines |
| Google Translation LLM | 0.9 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 1.9 | Open-weight LLMs |
| Amazon Translate | 1.9 | MT engines |
| Translated Lara 2 | 2 | LLM-based MT engines |
| Yandex Cloud Translate v2 | 3.4 | MT engines |
| SYSTRAN PNMT API | 5.1 | MT engines |
| Google Gemini 3.5 Flash | 14 | LLMs |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| Alibaba Machine Translation API | 16 | MT engines |
| xAI Grok 4.20 (non-reasoning) | 19 | LLMs |
| Cohere Tiny Aya | 20 | LLMs |
| OpenAI GPT-4.1 | 28 | LLMs |
| DeepSeek V3.2 | 39 | LLMs |
| Anthropic Claude Haiku 4.5 | 43 | LLMs |
| OpenAI GPT-5.6 Terra | 45 | LLMs |
| Meta Llama 4 Maverick | 46 | Open-weight LLMs |
| Mistral Medium 3.1 | 50 | LLMs |
| Swiss AI Apertus-8B-2509 | 51 | Open-weight LLMs |
| MiniMax M2.5 | 53 | LLMs |
| Tencent Cloud TMT | 54 | MT engines |
| Anthropic Claude Sonnet 5 | 56 | LLMs |
| Cohere Command A Translate | 57 | LLMs |
| xAI Grok 4.20 (reasoning) | 57 | LLMs |
| Mistral Large 3 | 62 | LLMs |
| Anthropic Claude Fable 5 | 66 | LLMs |
| Widn AI Tower Sugarloaf | 66 | LLM-based MT engines |
| Anthropic Claude Opus 4.8 | 69 | LLMs |
| AI21 Jamba Large 1.7 | 74 | LLMs |
| Google Gemini 3.1 Pro Preview | 79 | LLMs |
| Anthropic Claude Sonnet 4.6 | 89 | LLMs |
| AppTek NMT | 105 | MT engines |
| OpenAI GPT-5.5 | 112 | LLMs |
| OpenAI GPT-5.6 Sol | 131 | LLMs |
| Google Gemma 4 | 148 | Open-weight LLMs |
| z.ai GLM-5 | 268 | LLMs |
| MoonshotAI Kimi K2.5 | 274 | LLMs |
| Alibaba Qwen3-235B-A22B | 279 | Open-weight 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 Hindi.
Of the ten models with real-time translations into Hindi, nine are MT engines, three of them LLM-based, and one is the open-weight OpenAI gpt-oss-120b. None of the three real-time solutions on the shortlist meet general translation quality, tag handling or formatting, and translations from all three meet tone of voice. The solutions that meet every Basic or every Standard requirement take 66 to 178 seconds per 1,000 words, and the fastest under both is on Anthropic Claude Fable 5.
| 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 | 100% | ✓ | $1,045 | 228 |
| OpenAI GPT-5.6 Sol | three-step, with the gender-neutrality agent | 97.2% | ✗ | $983 | 268 |
For Hindi we check the formal register, with आप, and gender-neutral wording through neutral nouns and neutral rephrasing, never a default gender. With the gender-neutrality agent, the solution on OpenAI GPT-5.6 Sol meets every other Advanced requirement, with 4.0 weighted errors per 1,000 words and no major error. It misses only gender neutrality, where we require 100% of the sentences about people to be gender-neutral. The one on Anthropic Claude Fable 5 meets that bar, with 6.9 weighted errors per 1,000 words.
For the 13 shortlisted solutions in Hindi, the median share of sentences on which the three LQA judges agree is 96%. Both pairs of automatic metrics agree above their thresholds, and neither result is borderline. With agreement that high, the Intento LQA error counts on this page are reliable.
Leave feedback