The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Swedish 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 Swedish. |
| Tone of voice | The formal register 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 Swedish, 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.
The two solutions that meet every Standard requirement trade errors and time against cost. The one on OpenAI GPT-5.5 has 12.5 total errors per 1,000 words against 18.5 for the one on Google Gemini 3.1 Pro Preview, and takes 51 seconds per 1,000 words against 82. The one on Google Gemini 3.1 Pro Preview costs $4.23 per million characters against $10.13. So Optimize for selects OpenAI GPT-5.5 for quality and latency, and Google Gemini 3.1 Pro Preview for cost.
Among the solutions that meet a level, a cost cap of $4.22 per million characters or less leaves at most the one on OpenAI GPT-5.6 Terra, at $4.03, and it meets only the Basic requirements. A time limit of 50 seconds per 1,000 words or less leaves at most that same solution, at 24 seconds. A limit of one step leaves all three: each is a one-step solution without the gender-neutrality agent.
The Advanced requirements remove neither of the two solutions that meet the Standard requirements. Without the gender-neutrality agent, 100% of the sentences about people in the translations from OpenAI GPT-5.5 and from Google Gemini 3.1 Pro Preview are gender-neutral. So in Swedish, meeting the Advanced requirements costs as much and takes as long as meeting the Standard ones: Optimize for selects the same solutions.
OpenAI GPT-5.5
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
OpenAI GPT-5.5
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
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
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
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 Terra | 17.8 | $4.03 | 24 |
| Standard | Tier 1: simple solution + backup | ★★ | one-step | Google Gemini 3.1 Pro Preview | 18.5 | $4.23 | 82 |
| Advanced | Tier 1: simple solution + backup | ★★ | one-step | Google Gemini 3.1 Pro Preview | 18.5 | $4.23 | 82 |
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.
Of the three solutions that meet the Basic requirements, the Standard ones remove the one on OpenAI GPT-5.6 Terra, which misses formatting. The Advanced requirements keep the other two, on Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5, both in one step without the gender-neutrality agent. Under the Basic requirements, Optimize for selects OpenAI GPT-5.6 Terra for cost and latency and OpenAI GPT-5.5 for quality. Under the other two, it selects Google Gemini 3.1 Pro Preview for cost and OpenAI GPT-5.5 for quality and latency.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Mistral Large 3 | 0.71 | 33 | Misses a Standard requirement |
| z.ai GLM-5 | 1.78 | 31.6 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.81 | 37.2 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.13 | 26.4 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 4.03 | 17.8 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.23 | 18.5 | Meets every Standard requirement |
| Anthropic Claude Sonnet 5 | 5.3 | 32.6 | Misses a Standard requirement |
| Microsoft Translator | 10 | 61.6 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 10.13 | 12.5 | Meets every Standard requirement |
| Translated Lara 2 | 24.99 | 36.9 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 27 | Misses a Standard requirement |
None of the one-step solutions under $4.03 per million characters meet general translation quality. The four cheapest, on Mistral Large 3, z.ai GLM-5, OpenAI GPT-4.1 and Google Gemini 3.5 Flash, cost $0.71 to $3.13 per million characters and have 26.4 to 37.2 weighted errors per 1,000 words. The solutions on the MT engine Microsoft Translator and the two LLM-based MT engines cost $10 to $27.50 per million characters and have 27.0 to 61.6 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 31.6 |
| Tone of voice | 0 |
| Tag handling | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
Of tone of voice, tag handling, formatting and general translation quality, general translation quality is the one the fewest shortlisted one-step solutions meet, 3 out of 11. The other eight have 26.4 to 61.6 weighted errors per 1,000 words. None of the solutions on the MT engine Microsoft Translator and the two LLM-based MT engines meet general translation quality or formatting. The solutions on Mistral Large 3 and Anthropic Claude Sonnet 5 miss tag handling by small margins, with 99.6% and 99.2% of the tags right.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Swedish | Class |
|---|---|---|
| Microsoft Translator | 61.6 | MT engines |
| OpenAI GPT-4.1 | 37.2 | LLMs |
| Translated Lara 2 | 36.9 | LLM-based MT engines |
| Mistral Large 3 | 33 | LLMs |
| Anthropic Claude Sonnet 5 | 32.6 | LLMs |
| z.ai GLM-5 | 31.6 | LLMs |
| DeepL Next Gen | 27 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 26.4 | LLMs |
| Google Gemini 3.1 Pro Preview | 18.5 | LLMs |
| OpenAI GPT-5.6 Terra | 17.8 | LLMs |
| OpenAI GPT-5.5 | 12.5 | LLMs |
No translation into Swedish: 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).
By Intento LQA, translations from the eight shortlisted LLMs have 12.5 to 37.2 weighted errors per 1,000 words. Translations from the two LLM-based MT engines, DeepL Next Gen and Translated Lara 2, have 27.0 and 36.9 weighted errors per 1,000 words, and those from the MT engine Microsoft Translator 61.6. All three solutions that meet general translation quality are on LLMs. The shortlisted LLM that does not meet our language-support criteria is Google Gemini 3.5 Flash, whose translations have two critical errors.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| DeepSeek V3.2 | 0.807 | 9 | 1.15 | 12 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.807 | 10 | 1.21 | 17 | No | Not shortlisted |
| Cohere Command A Translate | 0.805 | 11 | 1.17 | 14 | No | Not shortlisted |
| Google Gemma 4 | 0.804 | 12 | 1.19 | 16 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.804 | 13 | 1.22 | 21 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.804 | 15 | 1.12 | 10 | No | Not shortlisted |
| Widn AI Tower Sugarloaf | 0.803 | 16 | 1.15 | 13 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.802 | 17 | 1.21 | 19 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.802 | 18 | 1.21 | 18 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.801 | 20 | 1.14 | 11 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.799 | 22 | 1.35 | 26 | No | Not shortlisted |
| Google Translation LLM | 0.798 | 23 | 1.61 | 32 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.798 | 24 | 1.29 | 25 | No | Not shortlisted |
| Tilde TildeOpen LLM engine | 0.798 | 25 | 1.27 | 24 | No | Not shortlisted |
| MiniMax M2.5 | 0.797 | 26 | 1.21 | 20 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.795 | 27 | 1.23 | 22 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.788 | 28 | 1.39 | 28 | No | Not shortlisted |
| Supertext LLM | 0.783 | 29 | 1.47 | 29 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.783 | 30 | 1.36 | 27 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.782 | 31 | 1.67 | 33 | No | Not shortlisted |
| Amazon Translate | 0.781 | 32 | 1.69 | 34 | No | Not shortlisted |
| NiuTrans | 0.778 | 33 | 1.60 | 31 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.778 | 34 | 1.51 | 30 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.774 | 35 | 2.04 | 36 | No | Not shortlisted |
| Oracle Cloud Language Service | 0.763 | 36 | 1.71 | 35 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.759 | 37 | 2.09 | 37 | No | Not shortlisted |
| AppTek NMT | 0.714 | 38 | 3.11 | 39 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.699 | 39 | 2.87 | 38 | No | Not shortlisted |
| Cohere Tiny Aya | 0.647 | 40 | 4.05 | 40 | No | Not shortlisted |
| Google Gemini 3.1 Pro Preview | 0.817 | 1 | 0.99 | 2 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.816 | 2 | 1.10 | 8 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | 0.813 | 3 | 1.05 | 5 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.812 | 4 | 0.98 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.812 | 5 | 1.02 | 3 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.810 | 6 | 1.11 | 9 | Yes | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | 0.810 | 7 | 1.05 | 6 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.807 | 8 | 1.09 | 7 | Yes | Top 8 by both metrics |
| Translated Lara 2 | 0.804 | 14 | 1.26 | 23 | Yes | Real-time place |
| Mistral Large 3 | 0.801 | 19 | 1.03 | 4 | Yes | Top 8 by MetricX |
| Microsoft Translator | 0.800 | 21 | 1.18 | 15 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 1 | 0.817 | 2 | 0.99 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 2 | 0.816 | 8 | 1.10 | Top 8 by both metrics |
| OpenAI GPT-5.6 Terra | Adjusted LLM | 3 | 0.813 | 5 | 1.05 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 4 | 0.812 | 1 | 0.98 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 5 | 0.812 | 3 | 1.02 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 6 | 0.810 | 9 | 1.11 | Top 8 by COMET-Kiwi |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 7 | 0.810 | 6 | 1.05 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 8 | 0.807 | 7 | 1.09 | Top 8 by both metrics |
| DeepSeek V3.2 | Adjusted LLM | 9 | 0.807 | 12 | 1.15 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 10 | 0.807 | 17 | 1.21 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 11 | 0.805 | 14 | 1.17 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 12 | 0.804 | 16 | 1.19 | Not shortlisted |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 13 | 0.804 | 21 | 1.22 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 14 | 0.804 | 23 | 1.26 | Real-time place |
| Anthropic Claude Fable 5 | Adjusted LLM | 15 | 0.804 | 10 | 1.12 | Not shortlisted |
| Widn AI Tower Sugarloaf | Adjusted NMT | 16 | 0.803 | 13 | 1.15 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 17 | 0.802 | 19 | 1.21 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 18 | 0.802 | 18 | 1.21 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 19 | 0.801 | 4 | 1.03 | Top 8 by MetricX |
| Mistral Medium 3.1 | Adjusted LLM | 20 | 0.801 | 11 | 1.14 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 21 | 0.800 | 15 | 1.18 | Real-time place |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 22 | 0.799 | 26 | 1.35 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 23 | 0.798 | 32 | 1.61 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 24 | 0.798 | 25 | 1.29 | Not shortlisted |
| Tilde TildeOpen LLM engine | Adjusted NMT | 25 | 0.798 | 24 | 1.27 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 26 | 0.797 | 20 | 1.21 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 27 | 0.795 | 22 | 1.23 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 28 | 0.788 | 28 | 1.39 | Not shortlisted |
| Supertext LLM | Off-the-shelf NMT | 29 | 0.783 | 29 | 1.47 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 30 | 0.783 | 27 | 1.36 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 31 | 0.782 | 33 | 1.67 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 32 | 0.781 | 34 | 1.69 | Not shortlisted |
| NiuTrans | Off-the-shelf NMT | 33 | 0.778 | 31 | 1.60 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 34 | 0.778 | 30 | 1.51 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 35 | 0.774 | 36 | 2.04 | Not shortlisted |
| Oracle Cloud Language Service | Off-the-shelf NMT | 36 | 0.763 | 35 | 1.71 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 37 | 0.759 | 37 | 2.09 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 38 | 0.714 | 39 | 3.11 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 39 | 0.699 | 38 | 2.87 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 40 | 0.647 | 40 | 4.05 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.812 | — | 1.03 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.805 | — | 1.09 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.802 | — | 1.24 | 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 40 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 MetricX and fourth on COMET-Kiwi, have 27.0 weighted errors per 1,000 words, and those from Google Gemini 3.5 Flash, second on COMET-Kiwi, 26.4. Those from OpenAI GPT-5.5, fifth on COMET-Kiwi and third on MetricX, have the fewest, 12.5 weighted errors per 1,000 words. Nearest the cut are DeepSeek V3.2, ninth on COMET-Kiwi, and Anthropic Claude Fable 5, 10th 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 | $4.03 |
| Standard | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $4.23 |
| Advanced | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $4.23 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Mistral Large 3 | $0.71 | LLMs |
| z.ai GLM-5 | $1.78 | LLMs |
| OpenAI GPT-4.1 | $2.81 | LLMs |
| Google Gemini 3.5 Flash | $3.13 | LLMs |
| OpenAI GPT-5.6 Terra | $4.03 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.23 | LLMs |
| Anthropic Claude Sonnet 5 | $5.3 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $10.13 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.441 | LLMs |
| Anthropic Claude Fable 5 | 0.44 | LLMs |
| Anthropic Claude Sonnet 5 | 0.44 | LLMs |
| Mistral Medium 3.1 | 0.369 | LLMs |
| Mistral Large 3 | 0.367 | LLMs |
| AI21 Jamba Large 1.7 | 0.364 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.339 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.336 | LLMs |
| Cohere Command A Translate | 0.336 | LLMs |
| MoonshotAI Kimi K2.5 | 0.336 | LLMs |
| Widn AI Tower Sugarloaf | 0.328 | LLM-based MT engines |
| DeepSeek V3.2 | 0.326 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.325 | Open-weight LLMs |
| z.ai GLM-5 | 0.31 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.295 | LLMs |
| Google Gemini 3.5 Flash | 0.294 | LLMs |
| Google Gemma 4 | 0.293 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 0.285 | LLMs |
| OpenAI GPT-4.1 | 0.284 | LLMs |
| OpenAI GPT-5.6 Luna | 0.284 | LLMs |
| OpenAI GPT-5.6 Terra | 0.284 | LLMs |
| OpenAI GPT-5.5 | 0.283 | LLMs |
| MiniMax M2.5 | 0.277 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.274 | LLMs |
| Cohere Tiny Aya | 0.272 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.272 | LLMs |
The tokenizers produce a median 0.32 tokens per character of Swedish, 1.2 times as many as English text. Tokens per character do not set the order of the costs here. The tokenizer of OpenAI GPT-5.5 produces 0.283 tokens per character and its solution costs $10.13 per million characters, against 0.44 and $5.30 for Anthropic Claude Sonnet 5. The cheapest solution costs $4.23 per million characters under both the Standard and the Advanced requirements, because the one-step solution on Google Gemini 3.1 Pro Preview meets both.
| 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 | 24 |
| Standard | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 51 |
| Advanced | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 51 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Microsoft Translator | 0.2 | MT engines |
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| Google Translation LLM | 0.5 | LLM-based MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Translated Lara 2 | 1.2 | LLM-based MT engines |
| Amazon Translate | 1.3 | MT engines |
| OpenAI gpt-oss-120b | 1.8 | Open-weight LLMs |
| SYSTRAN PNMT API | 1.9 | MT engines |
| Yandex Cloud Translate v2 | 3.2 | MT engines |
| Tilde TildeOpen LLM engine | 7.3 | LLM-based MT engines |
| OpenAI GPT-5.6 Luna | 12 | LLMs |
| OpenAI GPT-4.1 | 13 | LLMs |
| Cohere Tiny Aya | 13 | LLMs |
| Supertext LLM | 16 | LLM-based MT engines |
| xAI Grok 4.20 (non-reasoning) | 23 | LLMs |
| OpenAI GPT-5.6 Terra | 24 | LLMs |
| Alibaba Machine Translation API | 24 | MT engines |
| Anthropic Claude Haiku 4.5 | 27 | LLMs |
| Mistral Medium 3.1 | 33 | LLMs |
| Swiss AI Apertus-8B-2509 | 39 | Open-weight LLMs |
| MiniMax M2.5 | 39 | LLMs |
| Anthropic Claude Sonnet 5 | 41 | LLMs |
| DeepSeek V3.2 | 41 | LLMs |
| OpenAI GPT-5.6 Sol | 42 | LLMs |
| Oracle Cloud Language Service | 44 | MT engines |
| Cohere Command A Translate | 45 | LLMs |
| Widn AI Tower Sugarloaf | 47 | LLM-based MT engines |
| Mistral Large 3 | 48 | LLMs |
| AI21 Jamba Large 1.7 | 49 | LLMs |
| Anthropic Claude Fable 5 | 50 | LLMs |
| OpenAI GPT-5.5 | 51 | LLMs |
| Anthropic Claude Opus 4.8 | 53 | LLMs |
| Anthropic Claude Sonnet 4.6 | 53 | LLMs |
| NiuTrans | 54 | MT engines |
| Google Gemini 3.5 Flash | 59 | LLMs |
| xAI Grok 4.20 (reasoning) | 70 | LLMs |
| Google Gemini 3.1 Pro Preview | 82 | LLMs |
| AppTek NMT | 92 | MT engines |
| Alibaba Qwen3-235B-A22B | 120 | Open-weight LLMs |
| Google Gemma 4 | 141 | Open-weight LLMs |
| z.ai GLM-5 | 157 | LLMs |
| MoonshotAI Kimi K2.5 | 169 | 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 Swedish.
Eight of the nine real-time models in Swedish are MT engines or LLM-based MT engines. The ninth is the open-weight OpenAI gpt-oss-120b, which did not make the shortlist. The three real-time solutions on the shortlist are on DeepL Next Gen, Translated Lara 2 and Microsoft Translator, and their translations have 27.0, 36.9 and 61.6 weighted errors per 1,000 words. Among the LLMs that are not open-weight, the time ranges from 12 seconds per 1,000 words on OpenAI GPT-5.6 Luna to 169 on MoonshotAI Kimi K2.5.
| 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-5.5 | one-step | 100% | ✓ | $10.13 | 51 |
| Google Gemini 3.1 Pro Preview | one-step | 100% | ✓ | $4.23 | 82 |
For tone of voice in Swedish we check the formal register. In the translations from all 11 shortlisted one-step solutions, 100% of the sentences are in it. For gender neutrality we check that the translation uses hen or a neutral noun for a person of unknown gender, without a default gender. In Swedish we did not find a grammatical form that marks gender without a neutral alternative. Translations from Google Gemini 3.1 Pro Preview and OpenAI GPT-5.5 are gender-neutral without the gender-neutrality agent.
The three LQA judges agree on a median of 96% of the sentences of the 11 shortlisted solutions, and neither pair of automatic metrics is borderline in Swedish. So the Intento LQA results and the automatic scores on this page rest on measurement we can rely on.
Leave feedback