The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Finnish 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 Finnish. |
| Tone of voice | The formal register (te (teitittely)) 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 Finnish, 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. 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 solutions meet every Standard requirement in Finnish, and the one-step solution on OpenAI GPT-5.6 Sol has the fewest total errors, the lowest cost and the shortest time of them. The two backups are two-step solutions (translation + post-edit) that cost more and have more errors: the one on OpenAI GPT-5.5 has 18.5 total errors per 1,000 words at $41.87 per million characters, and the one on Google Gemini 3.1 Pro Preview 17.8 at $32.00. So Optimize for selects OpenAI GPT-5.6 Sol for quality, cost and latency alike.
Between the two backups, neither is ahead on all three measures. The two-step solution on Google Gemini 3.1 Pro Preview costs $32.00 per million characters, against $41.87 for the one on OpenAI GPT-5.5. Its translations have 17.8 total errors per 1,000 words against 18.5, but it takes 200 seconds per 1,000 words against 167.
Every solution that meets the Standard requirements in Finnish costs $7.59 per million characters or more and takes 66 seconds per 1,000 words or more, so a tighter cost cap or time limit rules them all out. A limit of one step leaves the solution on OpenAI GPT-5.6 Sol alone, with no backup. The three real-time solutions on the shortlist all miss general translation quality, tone of voice and tag handling.
The Advanced requirements do not add cost or time in Finnish. In the translations from both of the two best solutions, 100% of the sentences about people are gender-neutral without the gender-neutrality agent. So the one-step solution on OpenAI GPT-5.6 Sol meets them at the same 13.2 total errors per 1,000 words, $7.60 per million characters and 67 seconds per 1,000 words, with the two-step solution on Google Gemini 3.1 Pro Preview as its one backup.
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.6 Sol
one-step LLM adjusted for Standard requirements
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 | 13.2 | $7.60 | 67 |
| Standard | Tier 1: simple solution + backup | ★ ☆☆ | one-step | OpenAI GPT-5.6 Sol | 13.2 | $7.60 | 67 |
| Advanced | Tier 1: simple solution + backup | ★ ☆ | one-step | OpenAI GPT-5.6 Sol | 13.2 | $7.60 | 67 |
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.
In Finnish the three solutions that meet the Basic requirements also meet tone of voice, tag handling and formatting, so all three meet every Standard requirement. The Advanced requirements remove the two-step solution (translation + post-edit) on OpenAI GPT-5.5, since the Advanced requirements count only the two best solutions. A one-step solution meets every level without the gender-neutrality agent, and Optimize for selects the one on OpenAI GPT-5.6 Sol for quality, cost and latency at all three.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| xAI Grok 4.20 (reasoning) | 1.12 | 41.2 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.05 | 46.8 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.56 | 25.4 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.75 | 20.8 | Misses a Standard requirement |
| OpenAI GPT-5.6 Sol | 7.6 | 13.2 | Meets every Standard requirement |
| Widn AI Tower Sugarloaf | 10 | 81.7 | Misses a Standard requirement |
| Microsoft Translator | 10 | 96.6 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.05 | 21.8 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 15.81 | 20.4 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 33.9 | Misses a Standard requirement |
In Finnish, paying more than $7.60 per million characters does not come with fewer errors. The five one-step solutions that cost more than $7.60 per million characters have 20.4 to 96.6 weighted errors per 1,000 words, from Anthropic Claude Opus 4.8 at $15.81 to Microsoft Translator at $10.00, and the solution on DeepL Next Gen, the dearest at $27.50, has 33.9. The four cheaper ones, all on LLMs from $1.12 to $4.75 per million characters, have 20.8 to 46.8 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 33.9 |
| Tone of voice | 5.3 |
| Tag handling | 1.3 |
| Gender neutrality | 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 Finnish | Class |
|---|---|---|
| Microsoft Translator | 96.6 | MT engines |
| ModernMT Enterprise Edition | 83.1 | MT engines |
| Widn AI Tower Sugarloaf | 81.7 | LLM-based MT engines |
| OpenAI GPT-4.1 | 46.8 | LLMs |
| xAI Grok 4.20 (reasoning) | 41.2 | LLMs |
| DeepL Next Gen | 33.9 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 25.4 | LLMs |
| OpenAI GPT-5.5 | 21.8 | LLMs |
| Google Gemini 3.1 Pro Preview | 20.8 | LLMs |
| Anthropic Claude Opus 4.8 | 20.4 | LLMs |
| OpenAI GPT-5.6 Sol | 13.2 | LLMs |
No translation into Finnish: SpeakLeash Bielik-11B-v3.0-Instruct (wrong language), Widn AI Tower Anthill (cut off part-way, wrong language), Widn AI Tower Vesuvius (cut off part-way).
The five one-step solutions with the fewest weighted errors in Finnish are all on LLMs. Translations from the seven shortlisted LLMs have 13.2 to 46.8 weighted errors per 1,000 words, those from the LLM-based MT engines DeepL Next Gen and Widn AI Tower Sugarloaf 33.9 and 81.7, and those from the MT engines ModernMT Enterprise Edition and Microsoft Translator 83.1 and 96.6. All seven LLMs support Finnish by our measured criteria, and the one-step solution on one of them, OpenAI GPT-5.6 Sol, meets general translation quality.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Mistral Large 3 | 0.853 | 9 | 1.99 | 19 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.852 | 10 | 1.80 | 10 | No | Not shortlisted |
| DeepSeek V3.2 | 0.852 | 11 | 2.06 | 21 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.851 | 12 | 1.78 | 9 | No | Not shortlisted |
| z.ai GLM-5 | 0.851 | 13 | 2.10 | 24 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.851 | 14 | 1.98 | 18 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.849 | 16 | 1.96 | 16 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.849 | 17 | 1.90 | 14 | No | Not shortlisted |
| Translated Lara 2 | 0.848 | 20 | 1.88 | 13 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.845 | 21 | 1.97 | 17 | No | Not shortlisted |
| Google Gemma 4 | 0.842 | 22 | 2.08 | 23 | No | Not shortlisted |
| Pangeanic PangeaMT | 0.841 | 23 | 2.07 | 22 | No | Not shortlisted |
| MiniMax M2.5 | 0.839 | 24 | 2.31 | 28 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.835 | 25 | 2.27 | 27 | No | Not shortlisted |
| Supertext LLM | 0.835 | 26 | 2.01 | 20 | No | Not shortlisted |
| Google Translation LLM | 0.834 | 27 | 2.19 | 26 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.834 | 28 | 2.11 | 25 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.822 | 29 | 2.38 | 29 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.819 | 30 | 2.65 | 32 | No | Not shortlisted |
| Cohere Command A Translate | 0.819 | 31 | 2.44 | 30 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.811 | 32 | 2.82 | 33 | No | Not shortlisted |
| Amazon Translate | 0.810 | 33 | 3.06 | 36 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.809 | 34 | 3.02 | 35 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.803 | 35 | 2.90 | 34 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.798 | 36 | 2.52 | 31 | No | Not shortlisted |
| Cohere Tiny Aya | 0.766 | 37 | 3.13 | 37 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.765 | 38 | 3.22 | 38 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.720 | 39 | 4.75 | 39 | No | Not shortlisted |
| AppTek NMT | 0.549 | 40 | 9.17 | 40 | No | Not shortlisted |
| TartuNLP | 0.469 | 41 | 11.55 | 41 | No | Not shortlisted |
| OpenAI GPT-5.5 | 0.862 | 1 | 1.62 | 3 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.861 | 2 | 1.58 | 2 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.860 | 3 | 1.74 | 5 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | 0.856 | 4 | 1.57 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.855 | 5 | 1.75 | 7 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.853 | 6 | 1.74 | 6 | Yes | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | 0.853 | 7 | 1.91 | 15 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | 0.853 | 8 | 1.81 | 11 | Yes | Top 8 by COMET-Kiwi |
| Widn AI Tower Sugarloaf | 0.851 | 15 | 1.64 | 4 | Yes | Top 8 by MetricX |
| Microsoft Translator | 0.848 | 18 | 1.76 | 8 | Yes | Top 8 by MetricX |
| ModernMT Enterprise Edition | 0.848 | 19 | 1.81 | 12 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.5 | Adjusted LLM | 1 | 0.862 | 3 | 1.62 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 2 | 0.861 | 2 | 1.58 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 3 | 0.860 | 5 | 1.74 | Top 8 by both metrics |
| OpenAI GPT-5.6 Sol | Adjusted LLM | 4 | 0.856 | 1 | 1.57 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 5 | 0.855 | 7 | 1.75 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 6 | 0.853 | 6 | 1.74 | Top 8 by both metrics |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 7 | 0.853 | 15 | 1.91 | Top 8 by COMET-Kiwi |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 8 | 0.853 | 11 | 1.81 | Top 8 by COMET-Kiwi |
| Mistral Large 3 | Adjusted LLM | 9 | 0.853 | 19 | 1.99 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 10 | 0.852 | 10 | 1.80 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 11 | 0.852 | 21 | 2.06 | Not shortlisted |
| Tilde TildeOpen LLM engine | Off-the-shelf NMT | 12 | 0.851 | 9 | 1.78 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 13 | 0.851 | 24 | 2.10 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.851 | 18 | 1.98 | Not shortlisted |
| Widn AI Tower Sugarloaf | Adjusted NMT | 15 | 0.851 | 4 | 1.64 | Top 8 by MetricX |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 16 | 0.849 | 16 | 1.96 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 17 | 0.849 | 14 | 1.90 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 18 | 0.848 | 8 | 1.76 | Top 8 by MetricX |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 19 | 0.848 | 12 | 1.81 | Real-time place |
| Translated Lara 2 | Adjusted NMT | 20 | 0.848 | 13 | 1.88 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 21 | 0.845 | 17 | 1.97 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 22 | 0.842 | 23 | 2.08 | Not shortlisted |
| Pangeanic PangeaMT | Off-the-shelf NMT | 23 | 0.841 | 22 | 2.07 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 24 | 0.839 | 28 | 2.31 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 25 | 0.835 | 27 | 2.27 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 26 | 0.835 | 20 | 2.01 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 27 | 0.834 | 26 | 2.19 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 28 | 0.834 | 25 | 2.11 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 29 | 0.822 | 29 | 2.38 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 30 | 0.819 | 32 | 2.65 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 31 | 0.819 | 30 | 2.44 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 32 | 0.811 | 33 | 2.82 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 33 | 0.810 | 36 | 3.06 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 34 | 0.809 | 35 | 3.02 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 35 | 0.803 | 34 | 2.90 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 36 | 0.798 | 31 | 2.52 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 37 | 0.766 | 37 | 3.13 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 38 | 0.765 | 38 | 3.22 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 39 | 0.720 | 39 | 4.75 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 40 | 0.549 | 40 | 9.17 | Not shortlisted |
| TartuNLP | Off-the-shelf NMT | 41 | 0.469 | 41 | 11.55 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.852 | — | 1.70 | Same model family as OpenAI GPT-5.6 Sol |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.851 | — | 1.68 | Same model family as OpenAI GPT-5.6 Sol |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.841 | — | 2.02 | 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 41 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 Finnish shortlist differently. Translations from DeepL Next Gen, second on both metrics, have 33.9 weighted errors per 1,000 words, and the solution misses general translation quality, tone of voice, tag handling and formatting. Those from Anthropic Claude Opus 4.8, on the list through COMET-Kiwi alone, have 20.4 weighted errors per 1,000 words, the second fewest of the one-step solutions. Nearest the cut are Mistral Large 3, 9th on COMET-Kiwi, and Tilde TildeOpen LLM engine, 9th on MetricX.
| 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 | $7.60 |
| Standard | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | $7.60 |
| Advanced | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | $7.60 |
| Model | US dollars per million characters | Class |
|---|---|---|
| xAI Grok 4.20 (reasoning) | $1.12 | LLMs |
| OpenAI GPT-4.1 | $3.05 | LLMs |
| Google Gemini 3.5 Flash | $3.56 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.75 | LLMs |
| OpenAI GPT-5.6 Sol | $7.6 | LLMs |
| Widn AI Tower Sugarloaf | $10 | LLM-based MT engines |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $11.05 | LLMs |
| Anthropic Claude Opus 4.8 | $15.81 | LLMs |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.519 | LLMs |
| Anthropic Claude Sonnet 5 | 0.519 | LLMs |
| Anthropic Claude Fable 5 | 0.517 | LLMs |
| Mistral Medium 3.1 | 0.437 | LLMs |
| Mistral Large 3 | 0.432 | LLMs |
| AI21 Jamba Large 1.7 | 0.419 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.406 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.404 | LLMs |
| Widn AI Tower Sugarloaf | 0.383 | LLM-based MT engines |
| Cohere Command A Translate | 0.382 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.381 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 0.378 | LLMs |
| DeepSeek V3.2 | 0.367 | LLMs |
| z.ai GLM-5 | 0.358 | LLMs |
| Google Gemini 3.5 Flash | 0.334 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.333 | LLMs |
| Google Gemma 4 | 0.332 | Open-weight LLMs |
| MiniMax M2.5 | 0.313 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.309 | LLMs |
| OpenAI GPT-4.1 | 0.308 | LLMs |
| OpenAI GPT-5.6 Terra | 0.308 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.307 | LLMs |
| OpenAI GPT-5.5 | 0.307 | LLMs |
| OpenAI GPT-5.6 Luna | 0.307 | LLMs |
| OpenAI GPT-5.6 Sol | 0.306 | LLMs |
| Cohere Tiny Aya | 0.294 | 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 | 67 |
| Standard | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | 67 |
| Advanced | OpenAI GPT-5.6 Sol, one-step LLM adjusted for Standard requirements | 67 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.1 | MT engines |
| Microsoft Translator | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| Pangeanic PangeaMT | 0.6 | MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| Google Translation LLM | 0.7 | LLM-based MT engines |
| Translated Lara 2 | 1.3 | LLM-based MT engines |
| Amazon Translate | 1.3 | MT engines |
| OpenAI gpt-oss-120b | 2.2 | Open-weight LLMs |
| SYSTRAN PNMT API | 2.2 | MT engines |
| Oracle Cloud Language Service | 3.4 | MT engines |
| Yandex Cloud Translate v2 | 3.9 | MT engines |
| TartuNLP | 6.3 | MT engines |
| Tilde TildeOpen LLM engine | 6.8 | LLM-based MT engines |
| Supertext LLM | 16 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 16 | LLMs |
| OpenAI GPT-4.1 | 17 | LLMs |
| Cohere Tiny Aya | 18 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 21 | LLMs |
| Alibaba Machine Translation API | 26 | MT engines |
| OpenAI GPT-5.6 Terra | 33 | LLMs |
| Anthropic Claude Haiku 4.5 | 39 | LLMs |
| Mistral Medium 3.1 | 40 | LLMs |
| MiniMax M2.5 | 45 | LLMs |
| Swiss AI Apertus-8B-2509 | 51 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | 55 | LLMs |
| AI21 Jamba Large 1.7 | 56 | LLMs |
| Cohere Command A Translate | 59 | LLMs |
| Widn AI Tower Sugarloaf | 61 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 64 | LLMs |
| OpenAI GPT-5.6 Sol | 67 | LLMs |
| Mistral Large 3 | 67 | LLMs |
| Anthropic Claude Sonnet 5 | 71 | LLMs |
| Anthropic Claude Fable 5 | 71 | LLMs |
| OpenAI GPT-5.5 | 76 | LLMs |
| AppTek NMT | 82 | MT engines |
| Anthropic Claude Sonnet 4.6 | 83 | LLMs |
| Google Gemini 3.1 Pro Preview | 91 | LLMs |
| DeepSeek V3.2 | 98 | LLMs |
| Alibaba Qwen3-235B-A22B | 133 | Open-weight LLMs |
| Google Gemma 4 | 148 | Open-weight LLMs |
| Anthropic Claude Opus 4.8 | 165 | LLMs |
| z.ai GLM-5 | 189 | LLMs |
| MoonshotAI Kimi K2.5 | 218 | 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 Finnish.
The ten real-time models in Finnish are six MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b, and three of them are on the shortlist. OpenAI gpt-oss-120b takes 2.2 seconds per 1,000 words into Finnish, and it is not on the shortlist. The time does not change from one level to the next: the one-step solution on OpenAI GPT-5.6 Sol is the fastest that meets the Basic, Standard and Advanced requirements alike, at 67 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 |
|---|---|---|---|---|---|
| OpenAI GPT-5.6 Sol | one-step | 100% | ✓ | $7.60 | 67 |
| Google Gemini 3.1 Pro Preview | two-step | 100% | ✓ | $32 | 200 |
Finnish grammar has no gender for persons, so a translation is gender-neutral by default, apart from gendered terms, and the two best solutions meet gender neutrality without the gender-neutrality agent. Tone of voice splits the shortlist by model type: every one-step LLM solution has 100% of its sentences in the formal register, against 2.1% to 10.6% for the MT engines and LLM-based MT engines.
The three LQA judges agree on a median of 96% of the sentences of the 11 shortlisted solutions in Finnish. Each pair of automatic metrics agrees above its threshold even at the low end of its range, so neither result is borderline. The shortlist ranks and the Intento LQA error counts on this page rest on measurement we can rely on.
Leave feedback