The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Kannada 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 Kannada. |
| Tone of voice | The formal register (ನೀವು) 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 Kannada, 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.
Three three-step solutions (translation + LQA + repair) meet every Standard requirement in Kannada. The one on OpenAI GPT-5.6 Terra has the fewest total errors, 4.9 per 1,000 words, and the lowest cost, $570 per million characters, so optimizing for quality or cost selects it. Optimizing for latency selects the one on Google Gemini 3.1 Pro Preview, at 283 seconds per 1,000 words against 306 for OpenAI GPT-5.6 Terra. The one on OpenAI GPT-5.5 has 8.2 total errors per 1,000 words and costs $609 per million characters.
Without the gender-neutrality agent, only these three solutions meet every Basic requirement in Kannada. None of them cost less than $570 per million characters or take less than 282 seconds per 1,000 words. The two-step solutions (translation + post-edit) on OpenAI GPT-5.6 Terra, Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5 cost $12.84 to $42.64 per million characters and take 189 to 342 seconds per 1,000 words. They have 22.1 to 25.0 weighted errors per 1,000 words, over the bar of general translation quality.
The three-step solution (translation + LQA + repair) on OpenAI GPT-5.6 Terra meets every Advanced requirement without the gender-neutrality agent: 100% of the sentences about people in its translations are gender-neutral. The one on OpenAI GPT-5.5 has 99.1% without the agent and meets the Advanced requirements with it. The agent raises its cost from $609 to $999 per million characters, its time from 466 to 473 seconds per 1,000 words and its total errors from 8.2 to 11.9 per 1,000 words.
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
Google Gemini 3.1 Pro Preview
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
Google Gemini 3.1 Pro Preview
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-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 2: two or more solutions, none simple | ☆☆☆ | three-step | OpenAI GPT-5.6 Terra | 4.9 | $570 | 306 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆☆ | three-step | OpenAI GPT-5.6 Terra | 4.9 | $570 | 306 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | three-step | OpenAI GPT-5.6 Terra | 4.9 | $570 | 306 |
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 three-step solutions (translation + LQA + repair) on OpenAI GPT-5.6 Terra, Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5 meet every Basic and every Standard requirement in Kannada. Those on OpenAI GPT-5.6 Terra, without the gender-neutrality agent, and on OpenAI GPT-5.5, with it, also meet every Advanced requirement. Under the Advanced requirements, optimizing for quality, cost or latency selects the three-step solution on OpenAI GPT-5.6 Terra, without the agent.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.15 | 59.3 | Misses a Standard requirement |
| Mistral Large 3 | 1.92 | 56.4 | Misses a Standard requirement |
| xAI Grok 4.20 (reasoning) | 2.23 | 64.6 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.39 | 97.9 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.87 | 49.4 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 5.08 | 23.7 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.11 | 29 | Misses a Standard requirement |
| z.ai GLM-5 | 7 | 50.8 | Misses a Standard requirement |
| Microsoft Translator | 10 | 129.9 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 12.58 | 24.1 | Misses a Standard requirement |
| Google Translation LLM | 21.58 | 118 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 36.6 | Misses a Standard requirement |
Among the one-step solutions in Kannada, the two on LLM-based MT engines are the dearest: Translated Lara 2 at $24.99 per million characters and Google Translation LLM at $21.58, more than any on an LLM. Their translations have 36.6 and 118.0 weighted errors per 1,000 words, against 23.7 for the one on OpenAI GPT-5.6 Terra at $5.08. The one on the MT engine Microsoft Translator costs $10 per million characters, more than all on LLMs but the one on OpenAI GPT-5.5, and has 129.9 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 53.6 |
| Tag handling | 1.3 |
| Tone of voice | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Four one-step solutions in Kannada meet every Standard requirement but general translation quality: those on OpenAI GPT-5.6 Terra, OpenAI GPT-5.5, Google Gemini 3.1 Pro Preview and xAI Grok 4.20 (reasoning), with 23.7 to 64.6 weighted errors per 1,000 words. Three-step solutions (translation + LQA + repair) bring the first three under the bar, at 4.9 to 9.6 weighted errors per 1,000 words. The three-step solution on xAI Grok 4.20 (reasoning) has 27.0 weighted errors per 1,000 words and 7 major errors, over the bar.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Kannada | Class |
|---|---|---|
| Microsoft Translator | 129.9 | MT engines |
| Google Translation LLM | 118 | LLM-based MT engines |
| OpenAI GPT-4.1 | 97.9 | LLMs |
| xAI Grok 4.20 (reasoning) | 64.6 | LLMs |
| Google Gemma 4 | 59.3 | Open-weight LLMs |
| Mistral Large 3 | 56.4 | LLMs |
| z.ai GLM-5 | 50.8 | LLMs |
| Google Gemini 3.5 Flash | 49.4 | LLMs |
| Translated Lara 2 | 36.6 | LLM-based MT engines |
| Google Gemini 3.1 Pro Preview | 29 | LLMs |
| OpenAI GPT-5.5 | 24.1 | LLMs |
| OpenAI GPT-5.6 Terra | 23.7 | LLMs |
No translation into Kannada: AI21 Jamba Large 1.7 (cut off part-way).
In one step in Kannada, translations from the eight shortlisted LLMs that are not open-weight have 23.7 to 97.9 weighted errors per 1,000 words, those from the two shortlisted LLM-based MT engines, Translated Lara 2 and Google Translation LLM, 36.6 and 118.0, and those from the shortlisted MT engine Microsoft Translator 129.9.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| DeepSeek V3.2 | 0.719 | 10 | 1.39 | 10 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.717 | 12 | 1.44 | 12 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.709 | 13 | 1.57 | 13 | No | Not shortlisted |
| MiniMax M2.5 | 0.708 | 15 | 1.57 | 14 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.701 | 17 | 1.64 | 15 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.698 | 18 | 1.82 | 19 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.697 | 19 | 1.79 | 18 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.687 | 20 | 2.40 | 22 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.684 | 21 | 2.36 | 20 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.683 | 22 | 2.39 | 21 | No | Not shortlisted |
| Amazon Translate | 0.675 | 23 | 2.71 | 23 | No | Not shortlisted |
| NiuTrans | 0.645 | 24 | 2.83 | 24 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.542 | 25 | 7.26 | 25 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.519 | 26 | 7.82 | 26 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.517 | 27 | 7.94 | 28 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.515 | 28 | 7.91 | 27 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.508 | 29 | 8.45 | 30 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.448 | 30 | 8.36 | 29 | No | Not shortlisted |
| Cohere Command A Translate | 0.428 | 31 | 10.84 | 32 | No | Not shortlisted |
| AppTek NMT | 0.422 | 32 | 9.58 | 31 | No | Not shortlisted |
| OpenAI GPT-5.6 Terra | 0.731 | 1 | 1.25 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.727 | 2 | 1.18 | 3 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.727 | 3 | 1.30 | 7 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.726 | 4 | 1.16 | 2 | Yes | Top 8 by both metrics |
| Google Gemma 4 | 0.725 | 5 | 1.15 | 1 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.724 | 6 | 1.28 | 6 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | 0.723 | 7 | 1.39 | 11 | Yes | Top 8 by COMET-Kiwi |
| Mistral Large 3 | 0.722 | 8 | 1.26 | 5 | Yes | Top 8 by both metrics |
| Translated Lara 2 | 0.722 | 9 | 1.38 | 9 | Yes | Real-time place |
| Google Translation LLM | 0.718 | 11 | 1.70 | 17 | Yes | Real-time place |
| Microsoft Translator | 0.708 | 14 | 1.65 | 16 | Yes | Real-time place |
| OpenAI GPT-4.1 | 0.707 | 16 | 1.35 | 8 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Terra | Adjusted LLM | 1 | 0.731 | 4 | 1.25 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 2 | 0.727 | 3 | 1.18 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 3 | 0.727 | 7 | 1.30 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 4 | 0.726 | 2 | 1.16 | Top 8 by both metrics |
| Google Gemma 4 | Adjusted LLM | 5 | 0.725 | 1 | 1.15 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 6 | 0.724 | 6 | 1.28 | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 7 | 0.723 | 11 | 1.39 | Top 8 by COMET-Kiwi |
| Mistral Large 3 | Adjusted LLM | 8 | 0.722 | 5 | 1.26 | Top 8 by both metrics |
| Translated Lara 2 | Adjusted NMT | 9 | 0.722 | 9 | 1.38 | Real-time place |
| DeepSeek V3.2 | Adjusted LLM | 10 | 0.719 | 10 | 1.39 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 11 | 0.718 | 17 | 1.70 | Real-time place |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 12 | 0.717 | 12 | 1.44 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 13 | 0.709 | 13 | 1.57 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 14 | 0.708 | 16 | 1.65 | Real-time place |
| MiniMax M2.5 | Adjusted LLM | 15 | 0.708 | 14 | 1.57 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 16 | 0.707 | 8 | 1.35 | Top 8 by MetricX |
| OpenAI gpt-oss-120b | Adjusted LLM | 17 | 0.701 | 15 | 1.64 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 18 | 0.698 | 19 | 1.82 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 19 | 0.697 | 18 | 1.79 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 20 | 0.687 | 22 | 2.40 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 21 | 0.684 | 20 | 2.36 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 22 | 0.683 | 21 | 2.39 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 23 | 0.675 | 23 | 2.71 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 24 | 0.645 | 24 | 2.83 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 25 | 0.542 | 25 | 7.26 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 26 | 0.519 | 26 | 7.82 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 27 | 0.517 | 28 | 7.94 | Not shortlisted |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 28 | 0.515 | 27 | 7.91 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 29 | 0.508 | 30 | 8.45 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 30 | 0.448 | 29 | 8.36 | Not shortlisted |
| Cohere Command A Translate | Off-the-shelf LLM | 31 | 0.428 | 32 | 10.84 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 32 | 0.422 | 31 | 9.58 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.730 | — | 1.11 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.729 | — | 1.14 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.722 | — | 1.33 | 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 32 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 Google Gemini 3.5 Flash third in Kannada and Google Gemini 3.1 Pro Preview fourth, and MetricX ranks them seventh and second. By Intento LQA, the one-step solutions on them have 49.4 and 29.0 weighted errors per 1,000 words. xAI Grok 4.20 (reasoning) is on the shortlist through COMET-Kiwi alone, at rank 11 on MetricX, and OpenAI GPT-4.1 through MetricX alone, at rank 16 on COMET-Kiwi. Nearest the cut in the tail is DeepSeek V3.2, at rank 10 on both metrics.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, three-step | $570 |
| Standard | OpenAI GPT-5.6 Terra, three-step | $570 |
| Advanced | OpenAI GPT-5.6 Terra, three-step | $570 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.15 | Open-weight LLMs |
| Mistral Large 3 | $1.92 | LLMs |
| xAI Grok 4.20 (reasoning) | $2.23 | LLMs |
| OpenAI GPT-4.1 | $3.39 | LLMs |
| Google Gemini 3.5 Flash | $3.87 | LLMs |
| OpenAI GPT-5.6 Terra | $5.08 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.11 | LLMs |
| z.ai GLM-5 | $7 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $12.58 | LLMs |
| Google Translation LLM | $21.58 | LLM-based MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 1.464 | LLMs |
| Anthropic Claude Sonnet 4.6 | 1.455 | LLMs |
| Anthropic Claude Fable 5 | 1.44 | LLMs |
| Anthropic Claude Sonnet 5 | 1.44 | LLMs |
| z.ai GLM-5 | 1.414 | LLMs |
| Anthropic Claude Haiku 4.5 | 1.409 | LLMs |
| Cohere Command A Translate | 1.379 | LLMs |
| MiniMax M2.5 | 1.24 | LLMs |
| Alibaba Qwen3-235B-A22B | 1.139 | Open-weight LLMs |
| Mistral Large 3 | 1.119 | LLMs |
| Mistral Medium 3.1 | 1.117 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.716 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.716 | LLMs |
| DeepSeek V3.2 | 0.635 | LLMs |
| MoonshotAI Kimi K2.5 | 0.627 | LLMs |
| OpenAI GPT-4.1 | 0.362 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.361 | LLMs |
| Google Gemma 4 | 0.359 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 0.359 | LLMs |
| OpenAI GPT-5.6 Terra | 0.357 | LLMs |
| OpenAI GPT-5.6 Luna | 0.356 | LLMs |
| Google Gemini 3.5 Flash | 0.355 | LLMs |
| OpenAI GPT-5.5 | 0.355 | LLMs |
More tokens per character do not always mean a dearer one-step solution in Kannada. The tokenizer of Mistral Large 3 produces 1.119 tokens per character of the translation, against 0.357 for OpenAI GPT-5.6 Terra, and the one-step solution on Mistral Large 3 costs $1.92 per million characters, against $5.08. From the Standard to the Advanced requirements, the cost of the cheapest solution stays at $570 per million characters. It is the same three-step solution on OpenAI GPT-5.6 Terra, without the gender-neutrality agent.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, three-step | 283 |
| Standard | Google Gemini 3.1 Pro Preview, three-step | 283 |
| Advanced | OpenAI GPT-5.6 Terra, three-step | 306 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.1 | MT engines |
| Microsoft Translator | 0.3 | MT engines |
| ModernMT Enterprise Edition | 0.4 | MT engines |
| Google Translation LLM | 0.6 | LLM-based MT engines |
| Amazon Translate | 1.7 | MT engines |
| Translated Lara 2 | 2.3 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.9 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.5 | MT engines |
| OpenAI GPT-5.6 Luna | 17 | LLMs |
| Google Gemini 3.5 Flash | 18 | LLMs |
| Alibaba Machine Translation API | 28 | MT engines |
| OpenAI GPT-4.1 | 30 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 40 | LLMs |
| AppTek NMT | 48 | MT engines |
| Swiss AI Apertus-8B-2509 | 51 | Open-weight LLMs |
| NiuTrans | 60 | MT engines |
| DeepSeek V3.2 | 71 | LLMs |
| OpenAI GPT-5.6 Terra | 79 | LLMs |
| Mistral Medium 3.1 | 79 | LLMs |
| Anthropic Claude Haiku 4.5 | 81 | LLMs |
| Mistral Large 3 | 85 | LLMs |
| Google Gemini 3.1 Pro Preview | 87 | LLMs |
| Cohere Command A Translate | 90 | LLMs |
| xAI Grok 4.20 (reasoning) | 112 | LLMs |
| MiniMax M2.5 | 125 | LLMs |
| OpenAI GPT-5.6 Sol | 154 | LLMs |
| Anthropic Claude Sonnet 5 | 160 | LLMs |
| Anthropic Claude Sonnet 4.6 | 161 | LLMs |
| OpenAI GPT-5.5 | 162 | LLMs |
| Anthropic Claude Fable 5 | 162 | LLMs |
| Google Gemma 4 | 195 | Open-weight LLMs |
| Anthropic Claude Opus 4.8 | 210 | LLMs |
| MoonshotAI Kimi K2.5 | 219 | LLMs |
| z.ai GLM-5 | 422 | LLMs |
| Alibaba Qwen3-235B-A22B | 447 | 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 Kannada.
The seven real-time models in Kannada are four MT engines, two LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. Translations from the three on the shortlist meet tone of voice, and those from Translated Lara 2 also meet tag handling. None of the three meet general translation quality or formatting. Solutions that meet every Standard requirement take 283 to 466 seconds per 1,000 words, and the two that meet every Advanced requirement 306, on OpenAI GPT-5.6 Terra, and 473, on OpenAI GPT-5.5 with the gender-neutrality agent.
| 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 Terra | three-step | 100% | ✓ | $570 | 306 |
| OpenAI GPT-5.5 | three-step, with the gender-neutrality agent | 100% | ✓ | $999 | 473 |
For Kannada we check the formal register, with ನೀವು, and gender-neutral wording through the honorific plural and neutral nouns, with no default gender. The other options in Kannada are the gendered third-person singular, a neutral honorific plural and rephrasing. In the translations from both of the two best solutions, 100% of the sentences about people are gender-neutral: from the three-step solution on OpenAI GPT-5.6 Terra without the gender-neutrality agent, and from the one on OpenAI GPT-5.5 with it.
For the 12 shortlisted solutions in Kannada, the median share of sentences on which the three LQA judges agree is 93.0%. The agreement of both pairs of automatic metrics is clear of the threshold, not borderline. With agreement this high on both sides, the Intento LQA error counts and the shortlist ranks on this page carry weight.
Leave feedback