The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Odia (Oriya) 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 Odia (Oriya). |
| Tone of voice | The formal register (ଆପଣ) 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 Odia (Oriya), 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 11 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.
Two three-step solutions (translation + LQA + repair) meet every Standard requirement in Odia (Oriya), and the one on Google Gemini 3.1 Pro Preview has fewer errors, costs less and takes less time. It has 13.6 total errors per 1,000 words against 14.1 for the one on OpenAI GPT-5.5. It costs $647 against $703 per million characters and takes 372 against 896 seconds per 1,000 words. Optimizing for quality, cost or latency selects it.
None of the solutions that meet every Basic requirement in Odia (Oriya) cost less than $647 per million characters. None of the one-step or two-step solutions (translation + post-edit) meet even the Basic requirements, which cover general translation quality alone. The closest two-step solution, on Google Gemini 3.1 Pro Preview, has 32.0 weighted errors per 1,000 words and costs $63.30 per million characters.
For the Advanced requirements, the gender-neutrality agent works with both three-step solutions. On Google Gemini 3.1 Pro Preview the cost goes from $647 to $1,054 per million characters and the time from 372 to 360 seconds per 1,000 words. On OpenAI GPT-5.5 the cost goes from $703 to $1,173 per million characters and the time from 896 to 909 seconds per 1,000 words. Total errors fall from 13.6 to 12.9 per 1,000 words on Google Gemini 3.1 Pro Preview and rise from 14.1 to 15.5 on OpenAI GPT-5.5.
Google Gemini 3.1 Pro Preview
three-step
Google Gemini 3.1 Pro Preview
three-step
Google Gemini 3.1 Pro Preview
three-step
Google Gemini 3.1 Pro Preview
three-step
Google Gemini 3.1 Pro Preview
three-step
Google Gemini 3.1 Pro Preview
three-step
Google Gemini 3.1 Pro Preview
three-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
three-step with the gender-neutrality agent
Google Gemini 3.1 Pro Preview
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 2: two or more solutions, none simple | ☆☆ | three-step | Google Gemini 3.1 Pro Preview | 13.6 | $647 | 372 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | three-step | Google Gemini 3.1 Pro Preview | 13.6 | $647 | 372 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | with the gender-neutrality agent | Google Gemini 3.1 Pro Preview | 12.9 | $1,054 | 360 |
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 two three-step solutions (translation + LQA + repair) that meet every Basic requirement, on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5, also meet every Standard requirement. None of the one-step or two-step solutions (translation + post-edit) meet even the Basic requirements, so every level takes three steps. The Advanced level also takes the gender-neutrality agent, and with it both meet every Advanced requirement. Optimizing for quality, cost or latency selects the solution on Google Gemini 3.1 Pro Preview at every level.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.24 | 79.1 | Misses a Standard requirement |
| OpenAI GPT-5.6 Luna | 1.04 | 23.1 | Misses a Standard requirement |
| DeepSeek V3.2 | 2.24 | 68.9 | Misses a Standard requirement |
| xAI Grok 4.20 (reasoning) | 3.4 | 80.4 | Misses a Standard requirement |
| MoonshotAI Kimi K2.5 | 3.56 | 76.5 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 6.43 | 52.7 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 8.65 | 35.3 | Misses a Standard requirement |
| Microsoft Translator | 10 | 163.8 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 104.8 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 26.4 | 51.7 | Misses a Standard requirement |
Among the one-step solutions in Odia (Oriya), the one on OpenAI GPT-5.6 Luna has the fewest weighted errors, 23.1 per 1,000 words, and costs $1.04 per million characters. Only the one on Google Gemma 4 costs less, $0.24 per million characters, with 79.1 weighted errors per 1,000 words. The dearest, on OpenAI GPT-5.5 at $26.40 per million characters, has 51.7 weighted errors per 1,000 words, and the one on Translated Lara 2 costs $24.99 with 104.8.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 72.7 |
| Tag handling | 0.7 |
| Gender neutrality | 0.7 |
| Tone of voice | 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 Odia (Oriya) | Class |
|---|---|---|
| Microsoft Translator | 163.8 | MT engines |
| Translated Lara 2 | 104.8 | LLM-based MT engines |
| xAI Grok 4.20 (reasoning) | 80.4 | LLMs |
| Google Gemma 4 | 79.1 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 76.5 | LLMs |
| DeepSeek V3.2 | 68.9 | LLMs |
| Google Gemini 3.5 Flash | 52.7 | LLMs |
| OpenAI GPT-5.5 | 51.7 | LLMs |
| Google Gemini 3.1 Pro Preview | 35.3 | LLMs |
| OpenAI GPT-5.6 Luna | 23.1 | LLMs |
No translation into Odia (Oriya): AI21 Jamba Large 1.7 (garbage or near-empty, wrong language, wrong script, cut off part-way), Swiss AI Apertus-8B-2509 (cut off part-way, wrong language, wrong script, far too short).
Translations from the seven shortlisted LLMs other than the open-weight Google Gemma 4 meet formatting in one step. None of the one-step solutions meet it on the MT engine Microsoft Translator or on the LLM-based MT engines, Translated Lara 2 and Google Translation LLM. The four shortlisted LLMs outside our language-support criteria are DeepSeek V3.2, MoonshotAI Kimi K2.5, Google Gemma 4 and xAI Grok 4.20 (reasoning). Translations from them have 68.9 to 80.4 weighted errors per 1,000 words, over what the criteria allow.
| Rank | Model | COMET-Kiwi (higher is better) | 83% confidence interval, from | to | Status |
|---|---|---|---|---|---|
| 1 | Google Gemini 3.1 Pro Preview | 0.802 | 0.793 | 0.812 | Shortlisted |
| 2 | OpenAI GPT-5.6 Luna | 0.800 | 0.790 | 0.810 | Shortlisted |
| 3 | Google Gemini 3.5 Flash | 0.799 | 0.790 | 0.809 | Shortlisted |
| 4 | MoonshotAI Kimi K2.5 | 0.792 | 0.782 | 0.802 | Shortlisted |
| 5 | Google Gemma 4 | 0.791 | 0.781 | 0.801 | Shortlisted |
| 6 | Google Translation LLM | 0.791 | 0.777 | 0.805 | Shortlisted |
| 7 | DeepSeek V3.2 | 0.789 | 0.779 | 0.800 | Shortlisted |
| 8 | OpenAI GPT-5.5 | 0.788 | 0.778 | 0.797 | Shortlisted |
| 9 | Translated Lara 2 | 0.786 | 0.774 | 0.797 | Shortlisted |
| 10 | xAI Grok 4.20 (reasoning) | 0.786 | 0.775 | 0.796 | Shortlisted |
| 11 | z.ai GLM-5 | 0.785 | 0.775 | 0.796 | Not shortlisted |
| 12 | Microsoft Translator | 0.774 | 0.765 | 0.784 | Shortlisted |
| 13 | OpenAI GPT-4.1 | 0.766 | 0.755 | 0.777 | Not shortlisted |
| 14 | Alibaba Qwen3-235B-A22B | 0.761 | 0.748 | 0.773 | Not shortlisted |
| 15 | Google Cloud Advanced Translation API | 0.754 | 0.736 | 0.773 | Not shortlisted |
| 16 | OpenAI gpt-oss-120b | 0.754 | 0.742 | 0.766 | Not shortlisted |
| 17 | ModernMT Enterprise Edition | 0.746 | 0.731 | 0.762 | Not shortlisted |
| 18 | Mistral Large 3 | 0.744 | 0.731 | 0.758 | Not shortlisted |
| 19 | Mistral Medium 3.1 | 0.730 | 0.718 | 0.742 | Not shortlisted |
| 20 | Anthropic Claude Opus 4.8 | 0.657 | 0.625 | 0.690 | Not shortlisted |
| 21 | Anthropic Claude Sonnet 4.6 | 0.645 | 0.614 | 0.676 | Not shortlisted |
| 22 | Anthropic Claude Haiku 4.5 | 0.633 | 0.602 | 0.664 | Not shortlisted |
| 23 | Anthropic Claude Fable 5 | 0.628 | 0.594 | 0.662 | Not shortlisted |
| 24 | Anthropic Claude Sonnet 5 | 0.627 | 0.594 | 0.661 | Not shortlisted |
| 25 | MiniMax M2.5 | 0.611 | 0.598 | 0.625 | Not shortlisted |
| 26 | Cohere Command A Translate | 0.425 | 0.384 | 0.466 | Not shortlisted |
| 27 | Alibaba Machine Translation API | 0.338 | 0.318 | 0.359 | Not shortlisted |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | Shortlist |
|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 1 | 0.802 | Top 10 by COMET-Kiwi |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 2 | 0.800 | Top 10 by COMET-Kiwi |
| Google Gemini 3.5 Flash | Adjusted LLM | 3 | 0.799 | Top 10 by COMET-Kiwi |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 4 | 0.792 | Top 10 by COMET-Kiwi |
| Google Gemma 4 | Adjusted LLM | 5 | 0.791 | Top 10 by COMET-Kiwi |
| Google Translation LLM | Adjusted NMT | 6 | 0.791 | Top 10 by COMET-Kiwi |
| DeepSeek V3.2 | Adjusted LLM | 7 | 0.789 | Top 10 by COMET-Kiwi |
| OpenAI GPT-5.5 | Adjusted LLM | 8 | 0.788 | Top 10 by COMET-Kiwi |
| Translated Lara 2 | Adjusted NMT | 9 | 0.786 | Top 10 by COMET-Kiwi |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 10 | 0.786 | Top 10 by COMET-Kiwi |
| z.ai GLM-5 | Adjusted LLM | 11 | 0.785 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 12 | 0.774 | Real-time place |
| OpenAI GPT-4.1 | Adjusted LLM | 13 | 0.766 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 14 | 0.761 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 15 | 0.754 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 16 | 0.754 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 17 | 0.746 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 18 | 0.744 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 19 | 0.730 | Not shortlisted |
| Anthropic Claude Opus 4.8 | Off-the-shelf LLM | 20 | 0.657 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 21 | 0.645 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 22 | 0.633 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 23 | 0.628 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 24 | 0.627 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 25 | 0.611 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 26 | 0.425 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 27 | 0.338 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.794 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.796 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.785 | 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 27 models, with each model family counted once, and the top 10 in bold. Higher is better on COMET-Kiwi. The scores rank models within this language and are not a requirement result.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, three-step | $647 |
| Standard | Google Gemini 3.1 Pro Preview, three-step | $647 |
| Advanced | Google Gemini 3.1 Pro Preview, three-step with the gender-neutrality agent | $1,054 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.24 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | $1.04 | LLMs |
| DeepSeek V3.2 | $2.24 | LLMs |
| xAI Grok 4.20 (reasoning) | $3.4 | LLMs |
| MoonshotAI Kimi K2.5 | $3.56 | LLMs |
| Google Gemini 3.5 Flash | $6.43 | LLMs |
| Google Gemini 3.1 Pro Preview | $8.65 | LLMs |
| Microsoft Translator | $10 | MT engines |
| Google Translation LLM | $21.66 | LLM-based MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| OpenAI GPT-5.5 | $26.4 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Mistral Large 3 | 2.107 | LLMs |
| z.ai GLM-5 | 1.983 | LLMs |
| Mistral Medium 3.1 | 1.765 | LLMs |
| Cohere Command A Translate | 1.623 | LLMs |
| Alibaba Qwen3-235B-A22B | 1.593 | Open-weight LLMs |
| MiniMax M2.5 | 1.568 | LLMs |
| Anthropic Claude Fable 5 | 1.337 | LLMs |
| Anthropic Claude Opus 4.8 | 1.332 | LLMs |
| Anthropic Claude Sonnet 5 | 1.299 | LLMs |
| Anthropic Claude Haiku 4.5 | 1.284 | LLMs |
| Anthropic Claude Sonnet 4.6 | 1.277 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 1.157 | LLMs |
| xAI Grok 4.20 (reasoning) | 1.152 | LLMs |
| MoonshotAI Kimi K2.5 | 1.031 | LLMs |
| DeepSeek V3.2 | 0.936 | LLMs |
| OpenAI GPT-5.6 Sol | 0.816 | LLMs |
| OpenAI GPT-5.6 Luna | 0.814 | LLMs |
| OpenAI GPT-4.1 | 0.809 | LLMs |
| OpenAI GPT-5.5 | 0.808 | LLMs |
| OpenAI GPT-5.6 Terra | 0.804 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.617 | LLMs |
| Google Gemini 3.5 Flash | 0.601 | LLMs |
| Google Gemma 4 | 0.594 | Open-weight LLMs |
| 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 | 372 |
| Standard | Google Gemini 3.1 Pro Preview, three-step | 372 |
| Advanced | Google Gemini 3.1 Pro Preview, three-step with the gender-neutrality agent | 360 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.3 | MT engines |
| Google Cloud Advanced Translation API | 0.5 | MT engines |
| ModernMT Enterprise Edition | 0.5 | MT engines |
| Google Translation LLM | 1 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 5.4 | Open-weight LLMs |
| Alibaba Machine Translation API | 29 | MT engines |
| OpenAI GPT-5.6 Luna | 32 | LLMs |
| Translated Lara 2 | 38 | LLM-based MT engines |
| xAI Grok 4.20 (non-reasoning) | 54 | LLMs |
| DeepSeek V3.2 | 70 | LLMs |
| OpenAI GPT-4.1 | 70 | LLMs |
| Google Gemini 3.5 Flash | 75 | LLMs |
| Cohere Command A Translate | 79 | LLMs |
| Anthropic Claude Haiku 4.5 | 101 | LLMs |
| OpenAI GPT-5.6 Terra | 128 | LLMs |
| Google Gemini 3.1 Pro Preview | 129 | LLMs |
| xAI Grok 4.20 (reasoning) | 131 | LLMs |
| MiniMax M2.5 | 165 | LLMs |
| Mistral Medium 3.1 | 168 | LLMs |
| Anthropic Claude Sonnet 4.6 | 169 | LLMs |
| Anthropic Claude Sonnet 5 | 213 | LLMs |
| Anthropic Claude Fable 5 | 251 | LLMs |
| Anthropic Claude Opus 4.8 | 264 | LLMs |
| Mistral Large 3 | 279 | LLMs |
| Google Gemma 4 | 285 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 376 | LLMs |
| OpenAI GPT-5.5 | 379 | LLMs |
| OpenAI GPT-5.6 Sol | 387 | LLMs |
| Alibaba Qwen3-235B-A22B | 410 | Open-weight LLMs |
| z.ai GLM-5 | 2,639 | 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 Odia (Oriya).
The six real-time models in Odia (Oriya) are three MT engines, two LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. On the shortlist, translations from Microsoft Translator and Translated Lara 2 meet tone of voice and have 163.8 and 104.8 weighted errors per 1,000 words, with 3 and 1 critical errors. Those from Google Translation LLM have no conclusive Intento LQA result. The fastest solutions that meet every Standard and every Advanced requirement take 372 and 360 seconds per 1,000 words.
| The two best solutions | Solution | Sentences about people that are gender-neutral | Meets every Advanced requirement | Cost per million characters | Seconds per 1,000 words |
|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | three-step, with the gender-neutrality agent | 100% | ✓ | $1,054 | 360 |
| OpenAI GPT-5.5 | three-step, with the gender-neutrality agent | 100% | ✓ | $1,173 | 909 |
For Odia (Oriya) we check the formal register, with ଆପଣ, and gender-neutral wording. Odia (Oriya) has no grammatical gender for persons, so a translation is neutral by default as long as it has no gendered terms, and we allow no default gender. With the gender-neutrality agent, 100% of the sentences about people are gender-neutral in the translations from both of the two best solutions, the three-step ones on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5.
For the 11 shortlisted solutions in Odia (Oriya), the median share of sentences on which the three LQA judges agree is 93.8%. With agreement that high, the Intento LQA error counts on this page carry weight. The agreement of the similarity pair of automatic metrics in Odia (Oriya) is 0.65, above its threshold of 0.46.
Leave feedback