The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Czech 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 Czech. |
| Tone of voice | The formal register (vy (vykání)) 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 Czech, 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 one-step solutions meet every Standard requirement in Czech, and the one with the fewest errors is neither the cheapest nor the fastest. The solution on Anthropic Claude Fable 5 has 15.0 total errors per 1,000 words at $27.87 per million characters and 63 seconds per 1,000 words. The one on OpenAI GPT-5.6 Terra has 20.4 total errors per 1,000 words at $4.49 per million characters and 28 seconds per 1,000 words. Optimize for selects Anthropic Claude Fable 5 for quality and OpenAI GPT-5.6 Terra for cost and latency.
The third solution, on Anthropic Claude Opus 4.8, has 22.9 total errors per 1,000 words, costs $13.91 per million characters and takes 70 seconds per 1,000 words. A cost cap from $4.49 to $13.90 per million characters, or a time limit from 28 to just under 63 seconds per 1,000 words, leaves only the solution on OpenAI GPT-5.6 Terra. A cap of at most $4.48 per million characters or a limit of at most 27 seconds per 1,000 words removes all three. A limit of one step removes nothing, since all three are one-step solutions.
With the gender-neutrality agent, the solutions on Anthropic Claude Fable 5 and OpenAI GPT-5.6 Terra run as three-step solutions (translation + LQA + repair). These have 7.6 total errors per 1,000 words each, against 15.0 and 20.4 for their one-step solutions. They cost $1,032 and $977 per million characters, against $27.87 and $4.49 per million characters. They take 232 and 180 seconds per 1,000 words, against 63 and 28 seconds per 1,000 words.
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Terra
one-step LLM adjusted for Standard requirements
Anthropic Claude Fable 5
three-step with the gender-neutrality agent
Misses gender neutrality.
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 Terra | 20.4 | $4.49 | 28 |
| Standard | Tier 1: simple solution + backup | ★★★ | one-step | OpenAI GPT-5.6 Terra | 20.4 | $4.49 | 28 |
| Advanced | Tier 4: no solution | No solution | — | — | — |
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.
All three solutions that meet the Basic requirements also meet the Standard ones. The one-step solutions on Anthropic Claude Fable 5, OpenAI GPT-5.6 Terra and Anthropic Claude Opus 4.8 meet tone of voice, tag handling and formatting as well as general translation quality. The Advanced requirements remove all three, on gender neutrality. Optimize for picks the same solutions under the Basic and the Standard requirements, Anthropic Claude Fable 5 for quality and OpenAI GPT-5.6 Terra for cost and latency.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Mistral Large 3 | 0.81 | 34.3 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.09 | 31 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 4.49 | 15.2 | Meets every Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.62 | 20.8 | Misses a Standard requirement |
| Anthropic Claude Sonnet 5 | 5.53 | 35.3 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 6.63 | 22.1 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 13.91 | 18.1 | Meets every Standard requirement |
| Google Translation LLM | 20.33 | 59 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 35.9 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 29 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 27.87 | 11.2 | Meets every Standard requirement |
A higher price does not always come with fewer errors. The solution on Google Translation LLM costs $20.33 per million characters and has 59.0 weighted errors per 1,000 words, against $4.49 and 15.2 for OpenAI GPT-5.6 Terra. Of the solutions that meet every Standard requirement, the cost and errors front holds two LLMs, OpenAI GPT-5.6 Terra and Anthropic Claude Fable 5. The one on Anthropic Claude Opus 4.8 is off it, with 18.1 weighted errors per 1,000 words at $13.91 per million characters, against 15.2 at $4.49.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 30 |
| Gender neutrality | 5.2 |
| Tag handling | 0.2 |
| Tone of voice | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
The solution on Google Gemini 3.1 Pro Preview misses general translation quality by a small margin, with 20.8 weighted errors per 1,000 words, just over the bar. Translations from Anthropic Claude Sonnet 4.6, Mistral Large 3 and Cohere Command A Translate have 99.2% of the tags right. The six solutions that meet tag handling are all on LLMs. One step is enough here: with two-step solutions (translation + post-edit) or three-step solutions (translation + LQA + repair), Czech stays in tier 1 for the Standard requirements.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Czech | Class |
|---|---|---|
| Google Translation LLM | 59 | LLM-based MT engines |
| Cohere Command A Translate | 36.3 | LLMs |
| Translated Lara 2 | 35.9 | LLM-based MT engines |
| Anthropic Claude Sonnet 5 | 35.3 | LLMs |
| Mistral Large 3 | 34.3 | LLMs |
| OpenAI GPT-4.1 | 31 | LLMs |
| DeepL Next Gen | 29 | LLM-based MT engines |
| Anthropic Claude Sonnet 4.6 | 22.1 | LLMs |
| Google Gemini 3.1 Pro Preview | 20.8 | LLMs |
| Anthropic Claude Opus 4.8 | 18.1 | LLMs |
| OpenAI GPT-5.6 Terra | 15.2 | LLMs |
| Anthropic Claude Fable 5 | 11.2 | LLMs |
No translation into Czech: Widn AI Tower Anthill (cut off part-way, wrong language), Widn AI Tower Vesuvius (cut off part-way).
The nine shortlisted LLM solutions have from 11.2 to 36.3 weighted errors per 1,000 words, and the three LLM-based MT engines from 29.0 to 59.0. The three solutions that meet general translation quality are all on LLMs. Translations from Google Translation LLM are the only ones on the shortlist with critical errors, 7 of them. All nine shortlisted LLMs support Czech by the criteria we measure.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Google Gemini 3.5 Flash | 0.809 | 9 | 1.61 | 14 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.809 | 10 | 1.70 | 17 | No | Not shortlisted |
| OpenAI GPT-5.5 | 0.806 | 12 | 1.59 | 9 | No | Not shortlisted |
| z.ai GLM-5 | 0.806 | 13 | 1.61 | 13 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.804 | 14 | 1.65 | 16 | No | Not shortlisted |
| Widn AI Tower Sugarloaf | 0.804 | 15 | 1.62 | 15 | No | Not shortlisted |
| DeepSeek V3.2 | 0.803 | 16 | 1.60 | 10 | No | Not shortlisted |
| Google Gemma 4 | 0.799 | 20 | 1.72 | 18 | No | Not shortlisted |
| Microsoft Translator | 0.796 | 21 | 1.87 | 23 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.796 | 22 | 1.90 | 25 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.796 | 23 | 2.04 | 28 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.794 | 24 | 1.86 | 22 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.792 | 25 | 1.88 | 24 | No | Not shortlisted |
| MiniMax M2.5 | 0.792 | 26 | 2.07 | 29 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.791 | 27 | 1.84 | 20 | No | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | 0.788 | 28 | 1.93 | 26 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.787 | 29 | 2.16 | 31 | No | Not shortlisted |
| Supertext LLM | 0.785 | 30 | 1.84 | 21 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.780 | 31 | 2.08 | 30 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.780 | 32 | 2.53 | 36 | No | Not shortlisted |
| Amazon Translate | 0.776 | 33 | 2.32 | 33 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.776 | 34 | 2.31 | 32 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.773 | 35 | 2.35 | 34 | No | Not shortlisted |
| Cohere Tiny Aya | 0.752 | 36 | 2.39 | 35 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.752 | 37 | 2.89 | 37 | No | Not shortlisted |
| Baidu Translate API | 0.745 | 38 | 2.91 | 38 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.741 | 39 | 3.05 | 39 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.720 | 40 | 3.79 | 40 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.645 | 41 | 5.19 | 41 | No | Not shortlisted |
| AppTek NMT | 0.619 | 42 | 6.69 | 42 | No | Not shortlisted |
| DeepL Next Gen | 0.816 | 1 | 1.52 | 5 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | 0.814 | 2 | 1.44 | 1 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | 0.814 | 3 | 1.51 | 4 | Yes | Top 8 by both metrics |
| Cohere Command A Translate | 0.812 | 4 | 1.53 | 7 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.812 | 5 | 1.61 | 12 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | 0.812 | 6 | 1.47 | 2 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.810 | 7 | 1.52 | 6 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.810 | 8 | 1.61 | 11 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 4.6 | 0.808 | 11 | 1.50 | 3 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.803 | 17 | 1.78 | 19 | Yes | Real-time place |
| Mistral Large 3 | 0.803 | 18 | 1.54 | 8 | Yes | Top 8 by MetricX |
| Google Translation LLM | 0.799 | 19 | 1.97 | 27 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| DeepL Next Gen | Off-the-shelf NMT | 1 | 0.816 | 5 | 1.52 | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | Adjusted LLM | 2 | 0.814 | 1 | 1.44 | Top 8 by both metrics |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 3 | 0.814 | 4 | 1.51 | Top 8 by both metrics |
| Cohere Command A Translate | Adjusted LLM | 4 | 0.812 | 7 | 1.53 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 5 | 0.812 | 12 | 1.61 | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 6 | 0.812 | 2 | 1.47 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 7 | 0.810 | 6 | 1.52 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 8 | 0.810 | 11 | 1.61 | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | Adjusted LLM | 9 | 0.809 | 14 | 1.61 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 10 | 0.809 | 17 | 1.70 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 11 | 0.808 | 3 | 1.50 | Top 8 by MetricX |
| OpenAI GPT-5.5 | Adjusted LLM | 12 | 0.806 | 9 | 1.59 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 13 | 0.806 | 13 | 1.61 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.804 | 16 | 1.65 | Not shortlisted |
| Widn AI Tower Sugarloaf | Adjusted NMT | 15 | 0.804 | 15 | 1.62 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 16 | 0.803 | 10 | 1.60 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 17 | 0.803 | 19 | 1.78 | Real-time place |
| Mistral Large 3 | Adjusted LLM | 18 | 0.803 | 8 | 1.54 | Top 8 by MetricX |
| Google Translation LLM | Adjusted NMT | 19 | 0.799 | 27 | 1.97 | Real-time place |
| Google Gemma 4 | Adjusted LLM | 20 | 0.799 | 18 | 1.72 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 21 | 0.796 | 23 | 1.87 | Not shortlisted |
| Tilde TildeOpen LLM engine | Off-the-shelf NMT | 22 | 0.796 | 25 | 1.90 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.796 | 28 | 2.04 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 24 | 0.794 | 22 | 1.86 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 25 | 0.792 | 24 | 1.88 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 26 | 0.792 | 29 | 2.07 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 27 | 0.791 | 20 | 1.84 | Not shortlisted |
| SpeakLeash Bielik-11B-v3.0-Instruct | Adjusted LLM | 28 | 0.788 | 26 | 1.93 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 29 | 0.787 | 31 | 2.16 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 30 | 0.785 | 21 | 1.84 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 31 | 0.780 | 30 | 2.08 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 32 | 0.780 | 36 | 2.53 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 33 | 0.776 | 33 | 2.32 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 34 | 0.776 | 32 | 2.31 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 35 | 0.773 | 34 | 2.35 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 36 | 0.752 | 35 | 2.39 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 37 | 0.752 | 37 | 2.89 | Not shortlisted |
| Baidu Translate API | Off-the-shelf NMT | 38 | 0.745 | 38 | 2.91 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 39 | 0.741 | 39 | 3.05 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 40 | 0.720 | 40 | 3.79 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 41 | 0.645 | 41 | 5.19 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 42 | 0.619 | 42 | 6.69 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.806 | — | 1.52 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.809 | — | 1.45 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.805 | — | 1.72 | 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 42 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.
The metrics and Intento LQA order the shortlisted solutions differently. Translations from DeepL Next Gen, first on COMET-Kiwi, have 29.0 weighted errors per 1,000 words, against 15.2 for OpenAI GPT-5.6 Terra, first on MetricX. The solutions on Anthropic Claude Sonnet 4.6 and Mistral Large 3, on the list through MetricX alone, both miss general translation quality. Nearest the cut are Google Gemini 3.5 Flash, 9th on COMET-Kiwi, and OpenAI GPT-5.5, 9th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | $4.49 |
| Standard | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | $4.49 |
| Advanced | No solution | — |
| Model | US dollars per million characters | Class |
|---|---|---|
| Mistral Large 3 | $0.81 | LLMs |
| OpenAI GPT-4.1 | $3.09 | LLMs |
| OpenAI GPT-5.6 Terra | $4.49 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.62 | LLMs |
| Anthropic Claude Sonnet 5 | $5.53 | LLMs |
| Anthropic Claude Sonnet 4.6 | $6.63 | LLMs |
| Anthropic Claude Opus 4.8 | $13.91 | LLMs |
| Google Translation LLM | $20.33 | LLM-based MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $27.87 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Fable 5 | 0.472 | LLMs |
| Anthropic Claude Opus 4.8 | 0.472 | LLMs |
| Anthropic Claude Sonnet 5 | 0.472 | LLMs |
| AI21 Jamba Large 1.7 | 0.455 | LLMs |
| Mistral Medium 3.1 | 0.439 | LLMs |
| Mistral Large 3 | 0.436 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.421 | Open-weight LLMs |
| Widn AI Tower Sugarloaf | 0.421 | LLM-based MT engines |
| MoonshotAI Kimi K2.5 | 0.397 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.378 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.377 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.36 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.36 | LLMs |
| DeepSeek V3.2 | 0.356 | LLMs |
| z.ai GLM-5 | 0.355 | LLMs |
| MiniMax M2.5 | 0.343 | LLMs |
| OpenAI GPT-5.5 | 0.327 | LLMs |
| OpenAI GPT-4.1 | 0.326 | LLMs |
| OpenAI GPT-5.6 Terra | 0.326 | LLMs |
| OpenAI GPT-5.6 Luna | 0.324 | LLMs |
| OpenAI GPT-5.6 Sol | 0.324 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.322 | LLMs |
| Google Gemini 3.5 Flash | 0.321 | LLMs |
| Google Gemma 4 | 0.321 | Open-weight LLMs |
| Cohere Command A Translate | 0.302 | LLMs |
| Cohere Tiny Aya | 0.285 | LLMs |
The tokenizers produce a median of 0.36 tokens per character of the translation into Czech, 1.4 times their count for English. Tokens alone do not set the price. The tokenizers of Anthropic Claude Fable 5, Anthropic Claude Opus 4.8 and Anthropic Claude Sonnet 5 produce 0.472 tokens per character each, and their one-step solutions cost $27.87, $13.91 and $5.53 per million characters. The cheapest solution that meets every Basic requirement also meets every Standard one, so the Standard requirements add nothing to the cost.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | 28 |
| Standard | OpenAI GPT-5.6 Terra, one-step LLM adjusted for Standard requirements | 28 |
| Advanced | No solution | — |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| Google Cloud Advanced Translation API | 0.4 | MT engines |
| Pangeanic PangeaMT | 0.5 | MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Google Translation LLM | 0.7 | LLM-based MT engines |
| Translated Lara 2 | 1.3 | LLM-based MT engines |
| Amazon Translate | 1.4 | MT engines |
| SYSTRAN PNMT API | 1.5 | MT engines |
| OpenAI gpt-oss-120b | 2.1 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.1 | MT engines |
| Google Gemini 3.5 Flash | 13 | LLMs |
| Oracle Cloud Language Service | 13 | MT engines |
| OpenAI GPT-5.6 Luna | 14 | LLMs |
| Supertext LLM | 16 | LLM-based MT engines |
| Baidu Translate API | 16 | MT engines |
| xAI Grok 4.20 (non-reasoning) | 17 | LLMs |
| Cohere Tiny Aya | 18 | LLMs |
| OpenAI GPT-4.1 | 20 | LLMs |
| OpenAI GPT-5.6 Terra | 28 | LLMs |
| Cohere Command A Translate | 32 | LLMs |
| Anthropic Claude Haiku 4.5 | 34 | LLMs |
| MiniMax M2.5 | 37 | LLMs |
| xAI Grok 4.20 (reasoning) | 40 | LLMs |
| DeepSeek V3.2 | 43 | LLMs |
| Mistral Medium 3.1 | 47 | LLMs |
| Widn AI Tower Sugarloaf | 50 | LLM-based MT engines |
| OpenAI GPT-5.5 | 52 | LLMs |
| Alibaba Machine Translation API | 53 | MT engines |
| Anthropic Claude Sonnet 5 | 57 | LLMs |
| Anthropic Claude Sonnet 4.6 | 62 | LLMs |
| Anthropic Claude Fable 5 | 63 | LLMs |
| Mistral Large 3 | 63 | LLMs |
| SpeakLeash Bielik-11B-v3.0-Instruct | 68 | Open-weight LLMs |
| AI21 Jamba Large 1.7 | 69 | LLMs |
| OpenAI GPT-5.6 Sol | 69 | LLMs |
| Anthropic Claude Opus 4.8 | 70 | LLMs |
| Google Gemini 3.1 Pro Preview | 86 | LLMs |
| AppTek NMT | 96 | MT engines |
| Swiss AI Apertus-8B-2509 | 102 | Open-weight LLMs |
| Google Gemma 4 | 149 | Open-weight LLMs |
| Alibaba Qwen3-235B-A22B | 167 | Open-weight LLMs |
| Tilde TildeOpen LLM engine | 196 | LLM-based MT engines |
| z.ai GLM-5 | 232 | LLMs |
| MoonshotAI Kimi K2.5 | 254 | 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 Czech.
Six of the ten real-time models in Czech are MT engines, three are LLM-based MT engines and one is the open-weight OpenAI gpt-oss-120b. Of these, only the three LLM-based MT engines are on the shortlist. The shortlisted one-step LLM solutions take from 20 to 86 seconds per 1,000 words. Those that meet every Standard requirement take from 28 to 70 seconds per 1,000 words. The three-step solutions on OpenAI GPT-5.6 Terra and Anthropic Claude Fable 5 with the gender-neutrality agent take 180 and 232 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 |
|---|---|---|---|---|---|
| Anthropic Claude Fable 5 | three-step, with the gender-neutrality agent | 99.1% | ✗ | $1,032 | 232 |
| OpenAI GPT-5.6 Terra | three-step, with the gender-neutrality agent | 99.1% | ✗ | $977 | 180 |
For tone of voice we check the formal register, with vy (vykání) as the form of address. For gender neutrality we check neutral rephrasing plus impersonal forms. The other options in the language are the generic masculine, feminine pairing or slash forms, and neutral rephrasing. With the gender-neutrality agent, the two best solutions run as three-step solutions. Translations from these have 7.3 weighted errors per 1,000 words each, against 11.2 and 15.2 for the one-step solutions on Anthropic Claude Fable 5 and OpenAI GPT-5.6 Terra.
Leave feedback