The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Sundanese 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 Sundanese. |
| Tone of voice | The formal register (basa lemes) 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 Sundanese, 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.
The three-step solution (translation + LQA + repair) on Anthropic Claude Sonnet 4.6 has 24.1 total errors per 1,000 words, at $636 per million characters and 243 seconds per 1,000 words. The three-step one on Google Gemini 3.5 Flash has 28.6 total errors per 1,000 words, at $588 per million characters and 161 seconds per 1,000 words. The one-step solution on OpenAI GPT-5.6 Luna has 41.8 total errors per 1,000 words, at $0.41 per million characters and 14 seconds per 1,000 words. All three miss general translation quality.
A limit of one or two steps leaves the one-step solution on OpenAI GPT-5.5 with the fewest total errors, 36.3 per 1,000 words, at $10.29 per million characters. Of the solutions that cost less, the one-step solution on OpenAI GPT-5.6 Luna has the fewest, 41.8 total errors per 1,000 words, at $0.41 per million characters. Of the two real-time (<2 sec) solutions with a conclusive Intento LQA result, the one on DeepL Next Gen has fewer, 75.8 total errors per 1,000 words, against 109.5 for the one on OpenAI gpt-oss-120b.
The Advanced requirements add gender neutrality, and the two best solutions meet it without the gender-neutrality agent. So for the Advanced requirements, the three-step solution (translation + LQA + repair) on Anthropic Claude Sonnet 4.6 costs $636 per million characters and takes 243 seconds per 1,000 words, as for the Standard ones. It has 24.1 total errors per 1,000 words for both. The same holds for the two-step solution (translation + post-edit) on Anthropic Claude Opus 4.8, the other of the two best solutions.
Anthropic Claude Sonnet 4.6
three-step
Misses general translation quality.
Anthropic Claude Sonnet 4.6
three-step
Misses general translation quality.
Anthropic Claude Sonnet 4.6
three-step
Misses general translation quality.
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 4: no solution | No solution | — | — | — | ||
| Standard | Tier 4: no solution | No solution | — | — | — | ||
| 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.
In Sundanese, the Standard requirements add tone of voice, tags and formatting. Only two solutions meet all three and miss general translation quality alone: the three-step solution (translation + LQA + repair) on Anthropic Claude Sonnet 4.6 and the two-step solution (translation + post-edit) on Anthropic Claude Opus 4.8. The other three-step solutions miss tone of voice too. The Advanced requirements add gender neutrality, and translations from both have 100% of the sentences about people gender-neutral.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-5.6 Luna | 0.41 | 39.2 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 2.82 | 76.5 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.42 | 40.5 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 7.29 | 69.5 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 10.29 | 35.3 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 16.11 | 67.9 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 59.3 | Misses a Standard requirement |
Of the one-step solutions with a conclusive result and a cost, the two that cost more than the one on OpenAI GPT-5.5 have more weighted errors. The ones on Anthropic Claude Opus 4.8 and DeepL Next Gen cost $16.11 and $27.50 per million characters, and have 67.9 and 59.3 weighted errors per 1,000 words. The ones on Google Gemini 3.5 Flash, Anthropic Claude Sonnet 4.6 and OpenAI GPT-4.1 cost $2.82 to $7.29 per million characters and have 40.5 to 76.5 weighted errors per 1,000 words, all more than the cheaper one on OpenAI GPT-5.6 Luna.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 67.9 |
| Tone of voice | 3.9 |
| Tag handling | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
No solution meets general translation quality, with one, two or three steps. The three-step solutions (translation + LQA + repair) have 24.1 to 43.5 weighted errors per 1,000 words, all above the bar. The two-step solution (translation + post-edit) on OpenAI GPT-4.1 has 111.7 weighted errors per 1,000 words, against 76.5 with one step. Two solutions with more steps meet tone of voice: the three-step solution on Anthropic Claude Sonnet 4.6 and the two-step solution on Anthropic Claude Opus 4.8.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Sundanese | Class |
|---|---|---|
| Cohere Command A Translate | 519.4 | LLMs |
| OpenAI gpt-oss-120b | 97.9 | Open-weight LLMs |
| OpenAI GPT-4.1 | 76.5 | LLMs |
| Anthropic Claude Sonnet 4.6 | 69.5 | LLMs |
| Anthropic Claude Opus 4.8 | 67.9 | LLMs |
| DeepL Next Gen | 59.3 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 40.5 | LLMs |
| OpenAI GPT-5.6 Luna | 39.2 | LLMs |
| OpenAI GPT-5.5 | 35.3 | LLMs |
No translation into Sundanese: AI21 Jamba Large 1.7 (garbage or near-empty).
Translations from the two shortlisted LLMs with language support by our criteria, OpenAI GPT-5.5 and OpenAI GPT-5.6 Luna, have the fewest weighted errors of the one-step solutions. Translations from DeepL Next Gen, the one LLM-based MT engine with a conclusive Intento LQA result, have 59.3 weighted errors per 1,000 words. None of the three MT engines that are not LLM-based made the shortlist. The highest of them, Google Cloud Advanced Translation API, is at rank 23 on COMET-Kiwi and 22 on MetricX.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Mistral Large 3 | 0.684 | 12 | 2.11 | 18 | No | Not shortlisted |
| Google Gemma 4 | 0.682 | 13 | 1.91 | 14 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.680 | 15 | 2.13 | 19 | No | Not shortlisted |
| Google Gemini 3.1 Pro Preview | 0.678 | 16 | 1.83 | 10 | No | Not shortlisted |
| xAI Grok 4.20 (non-reasoning) | 0.677 | 17 | 1.96 | 16 | No | Not shortlisted |
| z.ai GLM-5 | 0.676 | 18 | 1.87 | 13 | No | Not shortlisted |
| DeepSeek V3.2 | 0.667 | 19 | 1.86 | 11 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.660 | 20 | 2.41 | 20 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.660 | 21 | 2.48 | 21 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.659 | 22 | 2.98 | 24 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.656 | 23 | 2.52 | 22 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.654 | 24 | 1.86 | 12 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.599 | 25 | 3.52 | 25 | No | Not shortlisted |
| MiniMax M2.5 | 0.588 | 26 | 5.23 | 27 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.575 | 27 | 4.26 | 26 | No | Not shortlisted |
| Cohere Command A Translate | 0.756 | 1 | 1.78 | 7 | Yes | Top 8 by both metrics |
| Anthropic Claude Haiku 4.5 | 0.736 | 2 | 1.93 | 15 | Yes | Top 8 by COMET-Kiwi |
| Swiss AI Apertus-8B-2509 | 0.700 | 3 | 2.66 | 23 | Yes | Top 8 by COMET-Kiwi |
| DeepL Next Gen | 0.699 | 4 | 1.54 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.694 | 5 | 1.64 | 5 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | 0.694 | 6 | 1.62 | 3 | Yes | Top 8 by both metrics |
| Translated Lara 2 | 0.693 | 7 | 1.82 | 9 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Luna | 0.692 | 8 | 1.59 | 2 | Yes | Top 8 by both metrics |
| OpenAI gpt-oss-120b | 0.688 | 9 | 1.96 | 17 | Yes | Real-time place |
| Google Gemini 3.5 Flash | 0.688 | 10 | 1.66 | 6 | Yes | Top 8 by MetricX |
| OpenAI GPT-5.5 | 0.684 | 11 | 1.63 | 4 | Yes | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | 0.682 | 14 | 1.80 | 8 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Cohere Command A Translate | Adjusted LLM | 1 | 0.756 | 7 | 1.78 | Top 8 by both metrics |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 2 | 0.736 | 15 | 1.93 | Top 8 by COMET-Kiwi |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 3 | 0.700 | 23 | 2.66 | Top 8 by COMET-Kiwi |
| DeepL Next Gen | Off-the-shelf NMT | 4 | 0.699 | 1 | 1.54 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 5 | 0.694 | 5 | 1.64 | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 6 | 0.694 | 3 | 1.62 | Top 8 by both metrics |
| Translated Lara 2 | Adjusted NMT | 7 | 0.693 | 9 | 1.82 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 8 | 0.692 | 2 | 1.59 | Top 8 by both metrics |
| OpenAI gpt-oss-120b | Off-the-shelf LLM | 9 | 0.688 | 17 | 1.96 | Real-time place |
| Google Gemini 3.5 Flash | Off-the-shelf LLM | 10 | 0.688 | 6 | 1.66 | Top 8 by MetricX |
| OpenAI GPT-5.5 | Adjusted LLM | 11 | 0.684 | 4 | 1.63 | Top 8 by MetricX |
| Mistral Large 3 | Adjusted LLM | 12 | 0.684 | 18 | 2.11 | Not shortlisted |
| Google Gemma 4 | Adjusted LLM | 13 | 0.682 | 14 | 1.91 | Not shortlisted |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 14 | 0.682 | 8 | 1.80 | Top 8 by MetricX |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 15 | 0.680 | 19 | 2.13 | Not shortlisted |
| Google Gemini 3.1 Pro Preview | Off-the-shelf LLM | 16 | 0.678 | 10 | 1.83 | Not shortlisted |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 17 | 0.677 | 16 | 1.96 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 18 | 0.676 | 13 | 1.87 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 19 | 0.667 | 11 | 1.86 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 20 | 0.660 | 20 | 2.41 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 21 | 0.660 | 21 | 2.48 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 22 | 0.659 | 24 | 2.98 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 23 | 0.656 | 22 | 2.52 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 24 | 0.654 | 12 | 1.86 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 25 | 0.599 | 25 | 3.52 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 26 | 0.588 | 27 | 5.23 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 27 | 0.575 | 26 | 4.26 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.676 | — | 1.53 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.677 | — | 1.71 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.674 | — | 1.78 | 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 27 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.
MetricX alone brought in the one-step solution with the fewest weighted errors, on OpenAI GPT-5.5, at rank 4 against rank 11 on COMET-Kiwi. The three solutions that came in through COMET-Kiwi alone, on Anthropic Claude Haiku 4.5, Swiss AI Apertus-8B-2509 and Translated Lara 2, have no conclusive Intento LQA result. Nearest the cut in the tail are Google Gemini 3.1 Pro Preview, at rank 10 on MetricX, and Mistral Large 3, at rank 12 on COMET-Kiwi.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | No solution | — |
| Standard | No solution | — |
| Advanced | No solution | — |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.41 | LLMs |
| Anthropic Claude Haiku 4.5 | $2.25 | LLMs |
| OpenAI GPT-4.1 | $2.82 | LLMs |
| Google Gemini 3.5 Flash | $3.42 | LLMs |
| Anthropic Claude Sonnet 4.6 | $7.29 | LLMs |
| OpenAI GPT-5.5 | $10.29 | LLMs |
| Anthropic Claude Opus 4.8 | $16.11 | 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 Fable 5 | 0.509 | LLMs |
| Anthropic Claude Opus 4.8 | 0.508 | LLMs |
| Anthropic Claude Sonnet 5 | 0.508 | LLMs |
| Mistral Large 3 | 0.407 | LLMs |
| Mistral Medium 3.1 | 0.407 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.386 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.377 | LLMs |
| MoonshotAI Kimi K2.5 | 0.355 | LLMs |
| z.ai GLM-5 | 0.351 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.35 | Open-weight LLMs |
| DeepSeek V3.2 | 0.344 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.325 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.325 | LLMs |
| MiniMax M2.5 | 0.317 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.316 | LLMs |
| Google Gemini 3.5 Flash | 0.31 | LLMs |
| Google Gemma 4 | 0.307 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 0.282 | LLMs |
| OpenAI GPT-5.6 Terra | 0.282 | LLMs |
| Cohere Command A Translate | 0.279 | LLMs |
| OpenAI GPT-5.6 Luna | 0.279 | LLMs |
| OpenAI GPT-5.5 | 0.277 | LLMs |
| OpenAI GPT-4.1 | 0.276 | LLMs |
The tokenizers of 23 models produce a median 0.33 tokens per character of the translation, 1.4 times as many as for English. Tokens per character do not set the order of the costs alone. With 0.277 tokens per character, the one-step solution on OpenAI GPT-5.5 costs $10.29 per million characters, and with 0.377, the one on Anthropic Claude Haiku 4.5 costs $2.25. On Anthropic Claude Sonnet 4.6, one step costs $7.29 per million characters, two steps (translation + post-edit) $16.16 and three steps (translation + LQA + repair) $636.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | No solution | — |
| Standard | No solution | — |
| Advanced | No solution | — |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.2 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Translated Lara 2 | 1.6 | LLM-based MT engines |
| OpenAI gpt-oss-120b | 2.1 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 3.3 | MT engines |
| OpenAI GPT-5.6 Luna | 14 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 17 | LLMs |
| Anthropic Claude Haiku 4.5 | 33 | LLMs |
| Cohere Command A Translate | 38 | LLMs |
| OpenAI GPT-4.1 | 38 | LLMs |
| Mistral Medium 3.1 | 40 | LLMs |
| Swiss AI Apertus-8B-2509 | 43 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | 44 | LLMs |
| OpenAI GPT-5.6 Terra | 45 | LLMs |
| MiniMax M2.5 | 50 | LLMs |
| Google Gemini 3.5 Flash | 55 | LLMs |
| DeepSeek V3.2 | 63 | LLMs |
| Mistral Large 3 | 66 | LLMs |
| Anthropic Claude Sonnet 5 | 71 | LLMs |
| Anthropic Claude Sonnet 4.6 | 79 | LLMs |
| Anthropic Claude Fable 5 | 79 | LLMs |
| OpenAI GPT-5.5 | 80 | LLMs |
| Google Gemini 3.1 Pro Preview | 81 | LLMs |
| Anthropic Claude Opus 4.8 | 89 | LLMs |
| OpenAI GPT-5.6 Sol | 91 | LLMs |
| Alibaba Qwen3-235B-A22B | 137 | Open-weight LLMs |
| Google Gemma 4 | 146 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 185 | LLMs |
| z.ai GLM-5 | 204 | 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 Sundanese.
Three of the five real-time models made the shortlist: the LLM-based MT engines DeepL Next Gen and Translated Lara 2, and the open-weight OpenAI gpt-oss-120b. Translations from DeepL Next Gen have 94.5% of the tags right. Of the shortlisted LLMs that are not open-weight, the one-step solution on OpenAI GPT-5.6 Luna is the fastest, at 14 seconds per 1,000 words. On Anthropic Claude Sonnet 4.6, one step takes 79 seconds per 1,000 words, two steps (translation + post-edit) 143 and three steps (translation + LQA + repair) 243.
| 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 Sonnet 4.6 | three-step | 100% | ✗ | $636 | 243 |
| Anthropic Claude Opus 4.8 | two-step | 100% | ✗ | $35.68 | 162 |
For tone of voice we check the formal register, with basa lemes as the form of address. With one step, translations have it in 29.8% to 95.7% of the sentences in the second person or the imperative. The two best solutions, the three-step solution (translation + LQA + repair) on Anthropic Claude Sonnet 4.6 and the two-step solution (translation + post-edit) on Anthropic Claude Opus 4.8, have it in 100%. Sundanese has no grammatical gender for persons, so a translation is gender-neutral unless it has a gendered term.
The metric agreement in Sundanese is not borderline. Across the 12 shortlisted solutions, the median agreement of the three LQA judges is 91.5% of the sentences. The general translation quality results on this page rest on the nine one-step solutions with a conclusive result, and on the two-step and three-step solutions on four LLMs among them: Anthropic Claude Sonnet 4.6, Anthropic Claude Opus 4.8, Google Gemini 3.5 Flash and OpenAI GPT-4.1.
Leave feedback