The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Tamil 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 Tamil. |
| Tone of voice | The formal register (அவர்) 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 Tamil, 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 12 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 two-step solutions (translation + post-edit) meet every Standard requirement in Tamil, and the one on OpenAI GPT-5.6 Sol has fewer total errors, a lower cost and a shorter time than the other. It has 10.5 total errors per 1,000 words, costs $9.10 per million characters and takes 200 seconds per 1,000 words. The one on Google Gemini 3.1 Pro Preview has 16.2 total errors per 1,000 words, costs $50.43 per million characters and takes 261 seconds per 1,000 words. Optimizing for quality, cost or latency selects OpenAI GPT-5.6 Sol.
A real-time limit (<2 sec) leaves three shortlisted solutions in Tamil, the one-step ones on DeepL Next Gen, Google Translation LLM and Translated Lara 2. None of them meet the Basic requirements. Of these, the one on DeepL Next Gen has the fewest weighted errors and the highest cost, $27.50 per million characters. Outside the limit, the one-step solution on Google Gemini 3.5 Flash takes 18 seconds per 1,000 words and has 33.3 weighted errors per 1,000 words, fewer than any of the three, at $3.41 per million characters.
The Advanced requirements add gender neutrality to the Standard ones. In Tamil the two-step solutions (translation + post-edit) on OpenAI GPT-5.6 Sol and Google Gemini 3.1 Pro Preview meet it without the gender-neutrality agent. So the Advanced requirements do not change the errors, cost or time of either solution. Optimizing for quality, cost or latency still selects OpenAI GPT-5.6 Sol, at $9.10 per million characters.
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
two-step
OpenAI GPT-5.6 Sol
two-step
OpenAI GPT-5.6 Sol
two-step
OpenAI GPT-5.6 Sol
two-step
OpenAI GPT-5.6 Sol
two-step
OpenAI GPT-5.6 Sol
two-step
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 | 10.8 | $9.10 | 200 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | two-step | OpenAI GPT-5.6 Sol | 10.5 | $9.10 | 200 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | two-step | OpenAI GPT-5.6 Sol | 10.5 | $9.10 | 200 |
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 Basic requirements cover general translation quality alone. In Tamil the one-step solution on OpenAI GPT-5.6 Sol and the two-step one (translation + post-edit) on Google Gemini 3.1 Pro Preview meet them. The Standard requirements remove the one-step solution on tag handling. The two-step solution on OpenAI GPT-5.6 Sol takes its place, with 10.5 total errors per 1,000 words against 10.8. The Advanced requirements remove neither two-step solution. At each level, optimizing for quality, cost or latency selects OpenAI GPT-5.6 Sol.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.14 | 65.6 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.41 | 33.3 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.53 | 21.8 | Misses a Standard requirement |
| z.ai GLM-5 | 6.08 | 44.2 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 9.1 | 10.5 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 9.5 | 52.4 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 13.14 | 34.3 | Misses a Standard requirement |
| Google Translation LLM | 22.82 | 108.8 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 23.95 | 33.6 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 47.1 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 43.2 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 47.93 | 23.7 | Misses a Standard requirement |
Among the one-step solutions in Tamil, the one on OpenAI GPT-5.6 Sol costs $9.10 per million characters and has the fewest weighted errors, 10.5 per 1,000 words. Every dearer one-step solution has more. The dearest, on Anthropic Claude Fable 5 at $47.93 per million characters, has 23.7 weighted errors per 1,000 words. The three on LLM-based MT engines cost $22.82 to $27.50 per million characters and have 43.2 to 108.8 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 38.8 |
| Tag handling | 1 |
| Tone of voice | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Tamil | Class |
|---|---|---|
| Google Translation LLM | 108.8 | LLM-based MT engines |
| Google Gemma 4 | 65.6 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 52.4 | LLMs |
| Translated Lara 2 | 47.1 | LLM-based MT engines |
| z.ai GLM-5 | 44.2 | LLMs |
| DeepL Next Gen | 43.2 | LLM-based MT engines |
| OpenAI GPT-5.5 | 34.3 | LLMs |
| Anthropic Claude Opus 4.8 | 33.6 | LLMs |
| Google Gemini 3.5 Flash | 33.3 | LLMs |
| Anthropic Claude Fable 5 | 23.7 | LLMs |
| Google Gemini 3.1 Pro Preview | 21.8 | LLMs |
| OpenAI GPT-5.6 Sol | 10.5 | LLMs |
No translation into Tamil: AI21 Jamba Large 1.7 (garbage or near-empty, cut off part-way).
Translations from the eight shortlisted LLMs that are not open-weight have 10.5 to 52.4 weighted errors per 1,000 words in one step in Tamil. Translations from Google Translation LLM have 11 critical errors, and those from the open-weight Google Gemma 4 have one. These are the only critical errors among the shortlisted one-step solutions. Google Gemma 4 is the shortlisted LLM that does not meet our language-support criteria, with 65.6 weighted errors per 1,000 words.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| DeepSeek V3.2 | 0.768 | 11 | 1.34 | 9 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.768 | 12 | 1.38 | 14 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.767 | 14 | 1.35 | 12 | No | Not shortlisted |
| Mistral Large 3 | 0.759 | 16 | 1.35 | 11 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.755 | 17 | 1.39 | 15 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.755 | 18 | 1.62 | 18 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.751 | 19 | 2.06 | 24 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.744 | 20 | 1.58 | 17 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.742 | 21 | 1.66 | 19 | No | Not shortlisted |
| Microsoft Translator | 0.739 | 22 | 1.82 | 20 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.731 | 23 | 1.92 | 22 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.730 | 24 | 1.94 | 23 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.725 | 25 | 2.61 | 31 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.721 | 26 | 1.90 | 21 | No | Not shortlisted |
| NiuTrans | 0.720 | 27 | 2.14 | 26 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.718 | 28 | 2.47 | 28 | No | Not shortlisted |
| MiniMax M2.5 | 0.708 | 29 | 2.52 | 29 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.705 | 30 | 2.40 | 27 | No | Not shortlisted |
| Amazon Translate | 0.705 | 31 | 2.58 | 30 | No | Not shortlisted |
| Cohere Tiny Aya | 0.697 | 32 | 2.95 | 32 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.634 | 33 | 4.17 | 33 | No | Not shortlisted |
| Cohere Command A Translate | 0.596 | 34 | 6.36 | 35 | No | Not shortlisted |
| AppTek NMT | 0.569 | 35 | 5.41 | 34 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.404 | 36 | 8.30 | 36 | No | Not shortlisted |
| OpenAI GPT-5.6 Sol | 0.787 | 1 | 1.22 | 4 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.782 | 2 | 1.12 | 1 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.782 | 3 | 1.14 | 2 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.778 | 4 | 1.20 | 3 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.778 | 5 | 1.36 | 13 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | 0.774 | 6 | 1.32 | 7 | Yes | Top 8 by both metrics |
| Google Gemma 4 | 0.773 | 7 | 1.35 | 10 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | 0.772 | 8 | 1.31 | 6 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.772 | 9 | 1.33 | 8 | Yes | Top 8 by MetricX |
| OpenAI GPT-5.5 | 0.771 | 10 | 1.25 | 5 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.767 | 13 | 1.44 | 16 | Yes | Real-time place |
| Google Translation LLM | 0.762 | 15 | 2.07 | 25 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Sol | Adjusted LLM | 1 | 0.787 | 4 | 1.22 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 2 | 0.782 | 1 | 1.12 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 3 | 0.782 | 2 | 1.14 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 4 | 0.778 | 3 | 1.20 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 5 | 0.778 | 13 | 1.36 | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 6 | 0.774 | 7 | 1.32 | Top 8 by both metrics |
| Google Gemma 4 | Adjusted LLM | 7 | 0.773 | 10 | 1.35 | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 8 | 0.772 | 6 | 1.31 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 9 | 0.772 | 8 | 1.33 | Top 8 by MetricX |
| OpenAI GPT-5.5 | Adjusted LLM | 10 | 0.771 | 5 | 1.25 | Top 8 by MetricX |
| DeepSeek V3.2 | Adjusted LLM | 11 | 0.768 | 9 | 1.34 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 12 | 0.768 | 14 | 1.38 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 13 | 0.767 | 16 | 1.44 | Real-time place |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 14 | 0.767 | 12 | 1.35 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 15 | 0.762 | 25 | 2.07 | Real-time place |
| Mistral Large 3 | Adjusted LLM | 16 | 0.759 | 11 | 1.35 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 17 | 0.755 | 15 | 1.39 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 18 | 0.755 | 18 | 1.62 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 19 | 0.751 | 24 | 2.06 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 20 | 0.744 | 17 | 1.58 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 21 | 0.742 | 19 | 1.66 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 22 | 0.739 | 20 | 1.82 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 23 | 0.731 | 22 | 1.92 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 24 | 0.730 | 23 | 1.94 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 25 | 0.725 | 31 | 2.61 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 26 | 0.721 | 21 | 1.90 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 27 | 0.720 | 26 | 2.14 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 28 | 0.718 | 28 | 2.47 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 29 | 0.708 | 29 | 2.52 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 30 | 0.705 | 27 | 2.40 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 31 | 0.705 | 30 | 2.58 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 32 | 0.697 | 32 | 2.95 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 33 | 0.634 | 33 | 4.17 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 34 | 0.596 | 35 | 6.36 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 35 | 0.569 | 34 | 5.41 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 36 | 0.404 | 36 | 8.30 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.785 | — | 1.13 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.779 | — | 1.23 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.768 | — | 1.48 | 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 36 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.
COMET-Kiwi ranks DeepL Next Gen fourth in Tamil and MetricX third. By Intento LQA, translations from it have 43.2 weighted errors per 1,000 words, more than those from six LLMs on the list. Google Gemini 3.5 Flash and Google Gemma 4 are on the list through COMET-Kiwi alone, and z.ai GLM-5 and OpenAI GPT-5.5 through MetricX alone. Their translations have 33.3 to 65.6 weighted errors per 1,000 words. Nearest the cut in the tail is DeepSeek V3.2, ninth on MetricX and 11th on COMET-Kiwi.
| 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 | $9.10 |
| Standard | OpenAI GPT-5.6 Sol, two-step | $9.10 |
| Advanced | OpenAI GPT-5.6 Sol, two-step | $9.10 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.14 | Open-weight LLMs |
| Google Gemini 3.5 Flash | $3.41 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.53 | LLMs |
| z.ai GLM-5 | $6.08 | LLMs |
| OpenAI GPT-5.6 Sol | $9.1 | LLMs |
| Anthropic Claude Sonnet 5 | $9.5 | LLMs |
| OpenAI GPT-5.5 | $13.14 | LLMs |
| Google Translation LLM | $22.82 | LLM-based MT engines |
| Anthropic Claude Opus 4.8 | $23.95 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $47.93 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| z.ai GLM-5 | 1.133 | LLMs |
| Cohere Command A Translate | 0.98 | LLMs |
| Mistral Large 3 | 0.977 | LLMs |
| Mistral Medium 3.1 | 0.972 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.923 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 0.764 | LLMs |
| Anthropic Claude Opus 4.8 | 0.754 | LLMs |
| Anthropic Claude Fable 5 | 0.751 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.748 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.72 | LLMs |
| MoonshotAI Kimi K2.5 | 0.628 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.503 | LLMs |
| MiniMax M2.5 | 0.498 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.496 | LLMs |
| DeepSeek V3.2 | 0.494 | LLMs |
| OpenAI GPT-5.6 Sol | 0.338 | LLMs |
| OpenAI GPT-5.6 Terra | 0.337 | LLMs |
| OpenAI GPT-5.6 Luna | 0.336 | LLMs |
| OpenAI GPT-4.1 | 0.334 | LLMs |
| OpenAI GPT-5.5 | 0.334 | LLMs |
| Google Gemma 4 | 0.275 | Open-weight LLMs |
| Cohere Tiny Aya | 0.274 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.274 | LLMs |
| Google Gemini 3.5 Flash | 0.274 | LLMs |
Tokens per character do not set the order of cost in Tamil. The tokenizers of the three shortlisted Anthropic Claude models produce 0.751 to 0.764 tokens per character of the translation, and the one-step solutions on them cost $9.50 to $47.93 per million characters. The tokenizer of z.ai GLM-5 produces 1.133 tokens per character, and its one-step solution costs $6.08 per million characters. On Google Gemini 3.1 Pro Preview, the one-step solution costs $4.53 per million characters and the two-step one (translation + post-edit) $50.43.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | 200 |
| Standard | OpenAI GPT-5.6 Sol, two-step | 200 |
| Advanced | OpenAI GPT-5.6 Sol, two-step | 200 |
| 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 | 0.4 | MT engines |
| Pangeanic PangeaMT | 0.5 | MT engines |
| Google Translation LLM | 0.6 | LLM-based MT engines |
| DeepL Next Gen | 0.8 | LLM-based MT engines |
| Amazon Translate | 1.3 | MT engines |
| SYSTRAN PNMT API | 1.6 | MT engines |
| Translated Lara 2 | 2.6 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 3.4 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 4.2 | MT engines |
| Google Gemini 3.5 Flash | 18 | LLMs |
| OpenAI GPT-5.6 Luna | 19 | LLMs |
| Cohere Tiny Aya | 20 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 26 | LLMs |
| Alibaba Machine Translation API | 32 | MT engines |
| OpenAI GPT-4.1 | 38 | LLMs |
| AppTek NMT | 40 | MT engines |
| DeepSeek V3.2 | 56 | LLMs |
| NiuTrans | 58 | MT engines |
| xAI Grok 4.20 (reasoning) | 64 | LLMs |
| Anthropic Claude Haiku 4.5 | 68 | LLMs |
| OpenAI GPT-5.6 Terra | 72 | LLMs |
| Google Gemini 3.1 Pro Preview | 77 | LLMs |
| Mistral Large 3 | 80 | LLMs |
| MiniMax M2.5 | 85 | LLMs |
| Mistral Medium 3.1 | 86 | LLMs |
| Cohere Command A Translate | 95 | LLMs |
| Anthropic Claude Sonnet 4.6 | 133 | LLMs |
| Anthropic Claude Fable 5 | 135 | LLMs |
| Anthropic Claude Sonnet 5 | 148 | LLMs |
| Swiss AI Apertus-8B-2509 | 152 | Open-weight LLMs |
| OpenAI GPT-5.5 | 159 | LLMs |
| Anthropic Claude Opus 4.8 | 171 | LLMs |
| Google Gemma 4 | 193 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 200 | LLMs |
| Alibaba Qwen3-235B-A22B | 369 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 404 | LLMs |
| z.ai GLM-5 | 537 | 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 Tamil.
The 10 real-time models in Tamil are six MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. The three LLM-based ones, DeepL Next Gen, Google Translation LLM and Translated Lara 2, are on the shortlist. The solutions on them meet tone of voice, not general translation quality or formatting, and translations from the first two have 94.1% and 95.1% of the tags right. The fastest solution that meets every Basic, Standard or Advanced requirement takes 200 seconds per 1,000 words, on OpenAI GPT-5.6 Sol.
| 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 |
|---|---|---|---|---|---|
| OpenAI GPT-5.6 Sol | two-step | 100% | ✓ | $9.10 | 200 |
| Google Gemini 3.1 Pro Preview | two-step | 100% | ✓ | $50.43 | 261 |
For Tamil we check the formal register, with அவர், and gender-neutral wording through the honorific plural or singular, neutral nouns and rephrasing, with no default gender. The two-step solutions (translation + post-edit) on OpenAI GPT-5.6 Sol and Google Gemini 3.1 Pro Preview have 100% of the sentences about people gender-neutral, without the gender-neutrality agent.
For the 12 shortlisted solutions in Tamil, the median share of sentences on which the three LQA judges agree is 95.5%. Both pairs of automatic metrics agree above the threshold, and neither pair is borderline. So the shortlist and the Intento LQA error counts on this page both rest on measures that agree.
Leave feedback