The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Gujarati 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 Gujarati. |
| Tone of voice | The formal register (આપ, તમે) 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 Gujarati, 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. Inconclusive: the three LQA judges agree on fewer than 90% of the sentences of the solution, so it has no general translation quality result (09-02). 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.
Of the two two-step solutions (translation + post-edit) that meet every Standard requirement in Gujarati, the one on Google Gemini 3.1 Pro Preview has fewer total errors, a lower cost and a shorter time than the one on OpenAI GPT-5.5. It has 17.9 against 20.9 total errors per 1,000 words. It costs $30.80 against $37.84 per million characters, and takes 206 against 281 seconds per 1,000 words. Optimizing for quality, cost or latency selects it.
Under the Basic requirements, the four solutions that meet them differ in errors, cost and time. The two-step solution on Google Gemini 3.1 Pro Preview has the fewest total errors, 17.9 per 1,000 words, at $30.80 per million characters and 206 seconds per 1,000 words. The one-step solution on OpenAI GPT-5.6 Sol has 23.6 total errors per 1,000 words, at $8.18 per million characters and 174 seconds per 1,000 words. Optimizing for quality selects the first, and optimizing for cost or latency the second.
None of the solutions that meet every Basic requirement in Gujarati cost less than $8.18 per million characters or take less than 173 seconds per 1,000 words. The one-step solution on OpenAI GPT-5.6 Sol sets both limits. The one real-time model on the shortlist is DeepL Next Gen. Its one-step solution does not meet even the Basic requirements, with 75.8 weighted errors per 1,000 words.
The gender-neutrality agent works with both two-step solutions that meet every Standard requirement. On Google Gemini 3.1 Pro Preview, the cost falls from $30.80 to $28.72 per million characters and the time from 206 to 200 seconds per 1,000 words. On OpenAI GPT-5.5, the cost rises from $37.84 to $48.88 per million characters and the time from 281 to 387 seconds per 1,000 words. On both, total errors fall from 17.9 to 16.5 and from 20.9 to 19.1 per 1,000 words, and weighted errors rise from 14.8 to 16.5 and from 18.5 to 18.8.
Google Gemini 3.1 Pro Preview
two-step
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
two-step
Google Gemini 3.1 Pro Preview
two-step
Google Gemini 3.1 Pro Preview
two-step
Google Gemini 3.1 Pro Preview
two-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
two-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
two-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 | 23.6 | $8.18 | 174 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | two-step | Google Gemini 3.1 Pro Preview | 17.9 | $30.80 | 206 |
| Advanced | Tier 3: one solution | ☆ | with the gender-neutrality agent | Google Gemini 3.1 Pro Preview | 16.5 | $28.72 | 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.
Four solutions meet every Basic requirement in Gujarati, two of them one-step, on OpenAI GPT-5.6 Sol and Anthropic Claude Fable 5. The Standard requirements remove both on tag handling and leave the two-step solutions (translation + post-edit) on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5. The Advanced requirements leave the one on Google Gemini 3.1 Pro Preview with the gender-neutrality agent. Under the Standard and Advanced requirements, optimizing for quality, cost or latency selects the one on Google Gemini 3.1 Pro Preview.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.15 | 45.2 | Misses a Standard requirement |
| DeepSeek V3.2 | 1.68 | 52.7 | Misses a Standard requirement |
| Mistral Large 3 | 2.58 | 53.7 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.84 | 41.2 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.21 | 21.4 | Misses a Standard requirement |
| z.ai GLM-5 | 6.19 | 63.9 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 8.18 | 19.8 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 8.68 | 44.8 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.7 | 21.4 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 22.03 | 23.1 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 75.8 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 44.78 | 18.8 | Misses a Standard requirement |
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 43 |
| Gender neutrality | 2.9 |
| Tag handling | 1.3 |
| Tone of voice | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
None of the one-step solutions in Gujarati meet both general translation quality and tag handling. Those on OpenAI GPT-5.6 Sol and Anthropic Claude Fable 5 meet the first, with 99.6% of the tags right. Three meet the second. Two of them, on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5, have 21.4 weighted errors per 1,000 words, too many for general translation quality. The two-step solutions (translation + post-edit) on these two models have 14.8 and 18.5 weighted errors per 1,000 words, with every tag right.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Gujarati | Class |
|---|---|---|
| DeepL Next Gen | 75.8 | LLM-based MT engines |
| z.ai GLM-5 | 63.9 | LLMs |
| Mistral Large 3 | 53.7 | LLMs |
| DeepSeek V3.2 | 52.7 | LLMs |
| Google Gemma 4 | 45.2 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 44.8 | LLMs |
| Google Gemini 3.5 Flash | 41.2 | LLMs |
| Anthropic Claude Opus 4.8 | 23.1 | LLMs |
| Google Gemini 3.1 Pro Preview | 21.4 | LLMs |
| OpenAI GPT-5.5 | 21.4 | LLMs |
| OpenAI GPT-5.6 Sol | 19.8 | LLMs |
| Anthropic Claude Fable 5 | 18.8 | LLMs |
No translation into Gujarati: AI21 Jamba Large 1.7 (cut off part-way, wrong language, wrong script).
All shortlisted solutions in Gujarati but one are on LLMs. In one step, translations from the open-weight Google Gemma 4 have 45.2 weighted errors per 1,000 words and 2 critical errors, the only critical errors among the shortlisted one-step solutions with an Intento LQA result. Translations from the other ten shortlisted LLMs with an Intento LQA result have 18.8 to 63.9 weighted errors per 1,000 words. DeepL Next Gen is the one LLM-based MT engine on the shortlist. None of the MT engines that are not LLM-based are on the shortlist.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| xAI Grok 4.20 (reasoning) | 0.684 | 10 | 1.45 | 10 | No | Not shortlisted |
| Microsoft Translator | 0.680 | 14 | 1.80 | 20 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.679 | 15 | 1.77 | 19 | No | Not shortlisted |
| Translated Lara 2 | 0.678 | 16 | 1.51 | 15 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.676 | 17 | 1.84 | 24 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.676 | 18 | 1.83 | 23 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.674 | 19 | 1.59 | 16 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.669 | 20 | 1.60 | 18 | No | Not shortlisted |
| Google Translation LLM | 0.658 | 21 | 1.80 | 21 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.657 | 22 | 2.39 | 28 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.656 | 23 | 1.60 | 17 | No | Not shortlisted |
| Amazon Translate | 0.656 | 24 | 2.64 | 29 | No | Not shortlisted |
| MiniMax M2.5 | 0.650 | 26 | 2.27 | 26 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.649 | 27 | 1.94 | 25 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.644 | 28 | 2.67 | 30 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.642 | 29 | 1.81 | 22 | No | Not shortlisted |
| Cohere Tiny Aya | 0.631 | 30 | 2.31 | 27 | No | Not shortlisted |
| AppTek NMT | 0.454 | 31 | 8.17 | 31 | No | Not shortlisted |
| Cohere Command A Translate | 0.419 | 32 | 10.39 | 32 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.697 | 1 | 1.14 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.694 | 2 | 1.26 | 2 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.692 | 3 | 1.51 | 14 | Yes | Top 8 by COMET-Kiwi |
| Google Gemma 4 | 0.689 | 4 | 1.42 | 9 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | 0.689 | 5 | 1.48 | 13 | Yes | Top 8 by COMET-Kiwi |
| z.ai GLM-5 | 0.688 | 6 | 1.47 | 11 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Sol | 0.687 | 7 | 1.32 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.686 | 8 | 1.48 | 12 | Yes | Top 8 by COMET-Kiwi |
| DeepSeek V3.2 | 0.686 | 9 | 1.37 | 6 | Yes | Top 8 by MetricX |
| DeepL Next Gen | 0.683 | 11 | 1.33 | 4 | Yes | Top 8 by MetricX |
| Mistral Large 3 | 0.682 | 12 | 1.35 | 5 | Yes | Top 8 by MetricX |
| Google Gemini 3.5 Flash | 0.681 | 13 | 1.41 | 7 | Yes | Top 8 by MetricX |
| Mistral Medium 3.1 | 0.656 | 25 | 1.42 | 8 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Anthropic Claude Fable 5 | Adjusted LLM | 1 | 0.697 | 1 | 1.14 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 2 | 0.694 | 2 | 1.26 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 3 | 0.692 | 14 | 1.51 | Top 8 by COMET-Kiwi |
| Google Gemma 4 | Adjusted LLM | 4 | 0.689 | 9 | 1.42 | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 5 | 0.689 | 13 | 1.48 | Top 8 by COMET-Kiwi |
| z.ai GLM-5 | Adjusted LLM | 6 | 0.688 | 11 | 1.47 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 7 | 0.687 | 3 | 1.32 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 8 | 0.686 | 12 | 1.48 | Top 8 by COMET-Kiwi |
| DeepSeek V3.2 | Adjusted LLM | 9 | 0.686 | 6 | 1.37 | Top 8 by MetricX |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 10 | 0.684 | 10 | 1.45 | Not shortlisted |
| DeepL Next Gen | Off-the-shelf NMT | 11 | 0.683 | 4 | 1.33 | Top 8 by MetricX |
| Mistral Large 3 | Adjusted LLM | 12 | 0.682 | 5 | 1.35 | Top 8 by MetricX |
| Google Gemini 3.5 Flash | Adjusted LLM | 13 | 0.681 | 7 | 1.41 | Top 8 by MetricX |
| Microsoft Translator | Off-the-shelf NMT | 14 | 0.680 | 20 | 1.80 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 15 | 0.679 | 19 | 1.77 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 16 | 0.678 | 15 | 1.51 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 17 | 0.676 | 24 | 1.84 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 18 | 0.676 | 23 | 1.83 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 19 | 0.674 | 16 | 1.59 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 20 | 0.669 | 18 | 1.60 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 21 | 0.658 | 21 | 1.80 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 22 | 0.657 | 28 | 2.39 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 23 | 0.656 | 17 | 1.60 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 24 | 0.656 | 29 | 2.64 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 25 | 0.656 | 8 | 1.42 | Top 8 by MetricX |
| MiniMax M2.5 | Adjusted LLM | 26 | 0.650 | 26 | 2.27 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 27 | 0.649 | 25 | 1.94 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 28 | 0.644 | 30 | 2.67 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 29 | 0.642 | 22 | 1.81 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 30 | 0.631 | 27 | 2.31 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 31 | 0.454 | 31 | 8.17 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 32 | 0.419 | 32 | 10.39 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.686 | — | 1.32 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.683 | — | 1.53 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.676 | — | 1.54 | 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 OpenAI GPT-5.5 third in Gujarati and MetricX 14th, so it is on the shortlist through COMET-Kiwi alone. By Intento LQA, its one-step solution has 21.4 weighted errors per 1,000 words, as many as the one on Google Gemini 3.1 Pro Preview, second on both metrics. MetricX ranks DeepL Next Gen fourth and COMET-Kiwi 11th, and its one-step solution has 75.8 weighted errors per 1,000 words, the most of the 12 with an Intento LQA result. Nearest the cut in the tail is xAI Grok 4.20 (reasoning), 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 Sol, one-step LLM adjusted for Standard requirements | $8.18 |
| Standard | Google Gemini 3.1 Pro Preview, two-step | $30.80 |
| Advanced | Google Gemini 3.1 Pro Preview, two-step with the gender-neutrality agent | $28.72 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.15 | Open-weight LLMs |
| DeepSeek V3.2 | $1.68 | LLMs |
| Mistral Large 3 | $2.58 | LLMs |
| Mistral Medium 3.1 | $3.31 | LLMs |
| Google Gemini 3.5 Flash | $3.84 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.21 | LLMs |
| z.ai GLM-5 | $6.19 | LLMs |
| OpenAI GPT-5.6 Sol | $8.18 | LLMs |
| Anthropic Claude Sonnet 5 | $8.68 | LLMs |
| OpenAI GPT-5.5 | $11.7 | LLMs |
| Anthropic Claude Opus 4.8 | $22.03 | LLMs |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $44.78 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Mistral Large 3 | 1.631 | LLMs |
| Mistral Medium 3.1 | 1.617 | LLMs |
| z.ai GLM-5 | 1.326 | LLMs |
| Cohere Command A Translate | 1.256 | LLMs |
| Alibaba Qwen3-235B-A22B | 1.181 | Open-weight LLMs |
| MiniMax M2.5 | 0.922 | LLMs |
| Anthropic Claude Opus 4.8 | 0.793 | LLMs |
| Anthropic Claude Fable 5 | 0.791 | LLMs |
| Anthropic Claude Sonnet 5 | 0.79 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.783 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.762 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.729 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.727 | LLMs |
| DeepSeek V3.2 | 0.707 | LLMs |
| MoonshotAI Kimi K2.5 | 0.681 | LLMs |
| Cohere Tiny Aya | 0.374 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.374 | LLMs |
| Google Gemini 3.5 Flash | 0.366 | LLMs |
| Google Gemma 4 | 0.366 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 0.352 | LLMs |
| OpenAI GPT-5.6 Terra | 0.351 | LLMs |
| OpenAI GPT-5.6 Luna | 0.348 | LLMs |
| OpenAI GPT-5.5 | 0.347 | LLMs |
| OpenAI GPT-4.1 | 0.345 | LLMs |
| 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 | 174 |
| Standard | Google Gemini 3.1 Pro Preview, two-step | 206 |
| Advanced | Google Gemini 3.1 Pro Preview, two-step with the gender-neutrality agent | 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 | 0.5 | MT engines |
| DeepL Next Gen | 1.1 | LLM-based MT engines |
| Google Translation LLM | 1.3 | LLM-based MT engines |
| Amazon Translate | 1.5 | MT engines |
| Translated Lara 2 | 2.5 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.6 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 5.2 | MT engines |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| Google Gemini 3.5 Flash | 16 | LLMs |
| OpenAI GPT-4.1 | 21 | LLMs |
| Cohere Tiny Aya | 25 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 31 | LLMs |
| AppTek NMT | 38 | MT engines |
| DeepSeek V3.2 | 60 | LLMs |
| OpenAI GPT-5.6 Terra | 66 | LLMs |
| Mistral Medium 3.1 | 71 | LLMs |
| xAI Grok 4.20 (reasoning) | 78 | LLMs |
| Cohere Command A Translate | 78 | LLMs |
| Anthropic Claude Haiku 4.5 | 85 | LLMs |
| Swiss AI Apertus-8B-2509 | 90 | Open-weight LLMs |
| MiniMax M2.5 | 91 | LLMs |
| Google Gemini 3.1 Pro Preview | 104 | LLMs |
| Mistral Large 3 | 124 | LLMs |
| OpenAI GPT-5.5 | 132 | LLMs |
| Anthropic Claude Sonnet 4.6 | 151 | LLMs |
| OpenAI GPT-5.6 Sol | 174 | LLMs |
| Google Gemma 4 | 177 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 232 | LLMs |
| Anthropic Claude Fable 5 | 309 | LLMs |
| Anthropic Claude Opus 4.8 | 311 | LLMs |
| MoonshotAI Kimi K2.5 | 320 | LLMs |
| Alibaba Qwen3-235B-A22B | 419 | Open-weight LLMs |
| z.ai GLM-5 | 526 | 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 Gujarati.
The real-time models in Gujarati are four MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. Only one of them, DeepL Next Gen, is on the shortlist. Translations from it meet tone of voice. They have 93.3% of the tags right and 75.8 weighted errors per 1,000 words. The solutions that meet every Basic requirement take 174 to 309 seconds per 1,000 words, and the one that meets every Advanced requirement 200.
| 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 |
|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | two-step, with the gender-neutrality agent | 100% | ✓ | $28.72 | 200 |
| OpenAI GPT-5.5 | two-step, with the gender-neutrality agent | 99.1% | ✗ | $48.88 | 387 |
For Gujarati we check the formal register, with આપ and તમે as forms of address, and gender-neutral wording through neutral nouns and neutral rephrasing, with no default gender. None of the one-step solutions meet gender neutrality. With the gender-neutrality agent, the two-step solution (translation + post-edit) on OpenAI GPT-5.5 meets every other Advanced requirement and misses only gender neutrality, where we require 100% of the sentences about people to be gender-neutral.
Leave feedback