The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Kazakh 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 Kazakh. |
| Tone of voice | The formal register (Сіз) 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 Kazakh, 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. 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 solutions meet every Standard requirement in Kazakh, and neither leads on errors, cost and time at once. The three-step solution (translation + LQA + repair) on Google Gemini 3.1 Pro Preview has 11.5 total errors per 1,000 words, against 19.1 for the two-step solution (translation + post-edit) on OpenAI GPT-5.5. It costs $591 against $11.19 per million characters and takes 314 against 108 seconds per 1,000 words. Optimizing for quality selects Google Gemini 3.1 Pro Preview, and for cost or latency OpenAI GPT-5.5.
A limit of one step leaves one solution that meets the Basic requirements in Kazakh, on OpenAI GPT-5.5. A cost cap from $11.20 to under $591 per million characters leaves only the solutions on OpenAI GPT-5.5 at every level. None of the solutions that meet the Basic requirements cost less than $11.19 per million characters or take less than 108 seconds per 1,000 words. None of the real-time solutions (<2 sec) have a conclusive Intento LQA result in Kazakh or meet even the Basic requirements.
The Advanced requirements add gender neutrality, and in Kazakh they do not change errors, cost or time. The three-step solution (translation + LQA + repair) on Google Gemini 3.1 Pro Preview and the two-step solution (translation + post-edit) on OpenAI GPT-5.5 have 100% of the sentences about people gender-neutral without the gender-neutrality agent. The cheapest solution costs $11.19 per million characters for the Standard and the Advanced requirements alike, and the one with the fewest errors $591.
Google Gemini 3.1 Pro Preview
three-step
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
three-step
OpenAI GPT-5.5
two-step
OpenAI GPT-5.5
two-step
Google Gemini 3.1 Pro Preview
three-step
OpenAI GPT-5.5
two-step
OpenAI GPT-5.5
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-5.5 | 19.1 | $11.19 | 108 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆ | two-step | OpenAI GPT-5.5 | 19.1 | $11.19 | 108 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | two-step | OpenAI GPT-5.5 | 19.1 | $11.19 | 108 |
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 in Kazakh cover general translation quality alone. Two solutions meet them: the one-step solution on OpenAI GPT-5.5 and the three-step solution (translation + LQA + repair) on Google Gemini 3.1 Pro Preview. The Standard requirements remove the one-step solution on formatting, and the two-step solution (translation + post-edit) on the same model meets them. The Advanced ones remove neither. At every level, optimizing for quality selects Google Gemini 3.1 Pro Preview, and for cost or latency OpenAI GPT-5.5.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-5.6 Luna | 0.44 | 32 | Misses a Standard requirement |
| xAI Grok 4.20 (non-reasoning) | 1.39 | 67.6 | Misses a Standard requirement |
| z.ai GLM-5 | 2.68 | 56.4 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 4.36 | 36.9 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.7 | 29 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 8.84 | 48.1 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.19 | 19.1 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 16.17 | 30.3 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 32.47 | 27.4 | Misses a Standard requirement |
Among the one-step solutions in Kazakh, the dearest is not the one with the fewest errors. The one on OpenAI GPT-5.5 has the fewest, 19.1 weighted errors per 1,000 words, at $11.19 per million characters. The dearest, on Anthropic Claude Fable 5, costs $32.47 per million characters and has 27.4 weighted errors per 1,000 words. The cheapest, on OpenAI GPT-5.6 Luna, costs $0.44 per million characters and has 32.0 weighted errors per 1,000 words, fewer than four dearer ones.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 32 |
| 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.
General translation quality is the requirement the most one-step solutions with a conclusive Intento LQA result miss in Kazakh. Eight of them have 27.4 to 67.6 weighted errors per 1,000 words, above the bar. Two-step solutions (translation + post-edit) on Google Gemini 3.1 Pro Preview and z.ai GLM-5 have 24.4 and 60.6 weighted errors per 1,000 words. Three-step solutions (translation + LQA + repair) on the same two models have 11.5 and 24.7 weighted errors per 1,000 words, and only the first meets the bar.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Kazakh | Class |
|---|---|---|
| xAI Grok 4.20 (non-reasoning) | 67.6 | LLMs |
| z.ai GLM-5 | 56.4 | LLMs |
| Anthropic Claude Sonnet 4.6 | 48.1 | LLMs |
| Google Gemini 3.5 Flash | 36.9 | LLMs |
| OpenAI GPT-5.6 Luna | 32 | LLMs |
| Anthropic Claude Opus 4.8 | 30.3 | LLMs |
| Google Gemini 3.1 Pro Preview | 29 | LLMs |
| Anthropic Claude Fable 5 | 27.4 | LLMs |
| OpenAI GPT-5.5 | 19.1 | LLMs |
All nine shortlisted solutions with a conclusive Intento LQA result in Kazakh are on LLMs, with 19.1 to 67.6 weighted errors per 1,000 words in one step. The two LLM-based MT engines on the shortlist, Translated Lara 2 and DeepL Next Gen, and the MT engine Microsoft Translator have no conclusive result. The LLM outside our language-support criteria is xAI Grok 4.20 (non-reasoning), with 67.6 weighted errors per 1,000 words against a limit of 60.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| OpenAI GPT-4.1 | 0.801 | 10 | 1.93 | 9 | No | Not shortlisted |
| DeepSeek V3.2 | 0.800 | 11 | 2.02 | 14 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.799 | 12 | 2.01 | 13 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.795 | 15 | 2.10 | 15 | No | Not shortlisted |
| Google Gemma 4 | 0.792 | 17 | 2.35 | 20 | No | Not shortlisted |
| Mistral Large 3 | 0.786 | 18 | 2.30 | 18 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.784 | 19 | 2.44 | 21 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.777 | 20 | 2.31 | 19 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.777 | 21 | 2.16 | 17 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.774 | 22 | 2.65 | 22 | No | Not shortlisted |
| Amazon Translate | 0.768 | 23 | 2.87 | 24 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.766 | 24 | 2.92 | 25 | No | Not shortlisted |
| MiniMax M2.5 | 0.763 | 25 | 2.98 | 26 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.763 | 26 | 2.80 | 23 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.756 | 27 | 3.15 | 28 | No | Not shortlisted |
| Cohere Tiny Aya | 0.741 | 28 | 3.15 | 27 | No | Not shortlisted |
| Cohere Command A Translate | 0.671 | 29 | 5.58 | 29 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.495 | 30 | 10.77 | 30 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.353 | 31 | 14.84 | 31 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.337 | 32 | 15.47 | 32 | No | Not shortlisted |
| OpenAI GPT-5.5 | 0.807 | 1 | 1.73 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.807 | 2 | 1.58 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.807 | 3 | 1.69 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.806 | 4 | 1.85 | 7 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.805 | 5 | 1.82 | 6 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.805 | 6 | 1.73 | 5 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.805 | 7 | 1.95 | 10 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 4.6 | 0.804 | 8 | 1.98 | 11 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | 0.802 | 9 | 1.65 | 2 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.799 | 13 | 1.98 | 12 | Yes | Real-time place |
| xAI Grok 4.20 (non-reasoning) | 0.799 | 14 | 1.90 | 8 | Yes | Top 8 by MetricX |
| Microsoft Translator | 0.792 | 16 | 2.14 | 16 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.5 | Adjusted LLM | 1 | 0.807 | 4 | 1.73 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 2 | 0.807 | 1 | 1.58 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 3 | 0.807 | 3 | 1.69 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 4 | 0.806 | 7 | 1.85 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 5 | 0.805 | 6 | 1.82 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 6 | 0.805 | 5 | 1.73 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 7 | 0.805 | 10 | 1.95 | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 8 | 0.804 | 11 | 1.98 | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | Adjusted LLM | 9 | 0.802 | 2 | 1.65 | Top 8 by MetricX |
| OpenAI GPT-4.1 | Adjusted LLM | 10 | 0.801 | 9 | 1.93 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 11 | 0.800 | 14 | 2.02 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 12 | 0.799 | 13 | 2.01 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 13 | 0.799 | 12 | 1.98 | Real-time place |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 14 | 0.799 | 8 | 1.90 | Top 8 by MetricX |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 15 | 0.795 | 15 | 2.10 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 16 | 0.792 | 16 | 2.14 | Real-time place |
| Google Gemma 4 | Adjusted LLM | 17 | 0.792 | 20 | 2.35 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 18 | 0.786 | 18 | 2.30 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 19 | 0.784 | 21 | 2.44 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 20 | 0.777 | 19 | 2.31 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 21 | 0.777 | 17 | 2.16 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 22 | 0.774 | 22 | 2.65 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 23 | 0.768 | 24 | 2.87 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 24 | 0.766 | 25 | 2.92 | Not shortlisted |
| MiniMax M2.5 | Off-the-shelf LLM | 25 | 0.763 | 26 | 2.98 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 26 | 0.763 | 23 | 2.80 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 27 | 0.756 | 28 | 3.15 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 28 | 0.741 | 27 | 3.15 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 29 | 0.671 | 29 | 5.58 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 30 | 0.495 | 30 | 10.77 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 31 | 0.353 | 31 | 14.84 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 32 | 0.337 | 32 | 15.47 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.802 | — | 1.63 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.802 | — | 1.89 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.797 | — | 2.17 | Same model family as xAI Grok 4.20 (non-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.
By Intento LQA, the one-step solution on OpenAI GPT-5.6 Luna, first on MetricX, has 32.0 weighted errors per 1,000 words, and the one on OpenAI GPT-5.5, fourth on MetricX, 19.1. Google Gemini 3.5 Flash and xAI Grok 4.20 (non-reasoning) are on the list through MetricX alone, z.ai GLM-5 and Anthropic Claude Sonnet 4.6 through COMET-Kiwi alone. The one-step solutions on these four have 36.9 to 67.6 weighted errors per 1,000 words. Nearest the cut in the tail is OpenAI GPT-4.1, at rank 9 on MetricX and rank 10 on COMET-Kiwi.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $11.19 |
| Standard | OpenAI GPT-5.5, two-step | $11.19 |
| Advanced | OpenAI GPT-5.5, two-step | $11.19 |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.44 | LLMs |
| xAI Grok 4.20 (non-reasoning) | $1.39 | LLMs |
| z.ai GLM-5 | $2.68 | LLMs |
| Google Gemini 3.5 Flash | $4.36 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.7 | LLMs |
| Anthropic Claude Sonnet 4.6 | $8.84 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $11.19 | LLMs |
| Anthropic Claude Opus 4.8 | $16.17 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $32.47 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| AI21 Jamba Large 1.7 | 0.839 | LLMs |
| Cohere Tiny Aya | 0.561 | LLMs |
| Mistral Medium 3.1 | 0.558 | LLMs |
| Mistral Large 3 | 0.551 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.547 | Open-weight LLMs |
| Anthropic Claude Sonnet 5 | 0.538 | LLMs |
| Anthropic Claude Fable 5 | 0.536 | LLMs |
| Anthropic Claude Opus 4.8 | 0.536 | LLMs |
| Cohere Command A Translate | 0.512 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.511 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.506 | LLMs |
| z.ai GLM-5 | 0.496 | LLMs |
| DeepSeek V3.2 | 0.495 | LLMs |
| MiniMax M2.5 | 0.431 | LLMs |
| MoonshotAI Kimi K2.5 | 0.421 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.406 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.404 | LLMs |
| Google Gemini 3.5 Flash | 0.398 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.396 | LLMs |
| Google Gemma 4 | 0.395 | Open-weight LLMs |
| OpenAI GPT-5.6 Terra | 0.31 | LLMs |
| OpenAI GPT-4.1 | 0.309 | LLMs |
| OpenAI GPT-5.6 Sol | 0.309 | LLMs |
| OpenAI GPT-5.5 | 0.307 | LLMs |
| OpenAI GPT-5.6 Luna | 0.307 | LLMs |
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 108 |
| Standard | OpenAI GPT-5.5, two-step | 108 |
| Advanced | OpenAI GPT-5.5, two-step | 108 |
| 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 | 0.9 | LLM-based MT engines |
| SYSTRAN PNMT API | 1.1 | MT engines |
| Translated Lara 2 | 1.8 | LLM-based MT engines |
| Amazon Translate | 1.8 | MT engines |
| OpenAI gpt-oss-120b | 2.4 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | 16 | LLMs |
| Google Gemini 3.5 Flash | 19 | LLMs |
| OpenAI GPT-4.1 | 19 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 27 | LLMs |
| Alibaba Machine Translation API | 30 | MT engines |
| Cohere Tiny Aya | 40 | LLMs |
| DeepSeek V3.2 | 42 | LLMs |
| OpenAI GPT-5.6 Terra | 48 | LLMs |
| Mistral Medium 3.1 | 52 | LLMs |
| Anthropic Claude Haiku 4.5 | 55 | LLMs |
| MiniMax M2.5 | 56 | LLMs |
| Swiss AI Apertus-8B-2509 | 60 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | 79 | LLMs |
| Mistral Large 3 | 80 | LLMs |
| AI21 Jamba Large 1.7 | 88 | LLMs |
| Cohere Command A Translate | 92 | LLMs |
| Anthropic Claude Sonnet 5 | 96 | LLMs |
| Google Gemini 3.1 Pro Preview | 103 | LLMs |
| OpenAI GPT-5.6 Sol | 106 | LLMs |
| OpenAI GPT-5.5 | 108 | LLMs |
| Anthropic Claude Fable 5 | 109 | LLMs |
| Anthropic Claude Sonnet 4.6 | 109 | LLMs |
| Anthropic Claude Opus 4.8 | 130 | LLMs |
| Google Gemma 4 | 159 | Open-weight LLMs |
| Alibaba Qwen3-235B-A22B | 192 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 193 | LLMs |
| z.ai GLM-5 | 302 | 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 Kazakh.
The real-time models in Kazakh are five MT engines, two LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. None of the three on the shortlist have a conclusive Intento LQA result or meet tag handling. The fastest one-step solution with a result, on OpenAI GPT-5.6 Luna, takes 16 seconds per 1,000 words and misses general translation quality. At every level, the fastest solution that meets the requirements takes 108 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 | 100% | ✓ | $591 | 314 |
| OpenAI GPT-5.5 | two-step | 100% | ✓ | $11.19 | 108 |
Translations from the MT engine Microsoft Translator have 91.5% of the sentences in the formal register in Kazakh, and those from the other 11 shortlisted one-step solutions, on LLMs and LLM-based MT engines, 100%.
For the 12 shortlisted solutions in Kazakh, the median share of sentences on which the three LQA judges agree is 92.7%. With agreement at that level, the Intento LQA error counts on this page carry weight. The agreement of each pair of automatic metrics is above its threshold and not borderline, so the ranking behind the shortlist carries weight too.
Leave feedback