The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Armenian 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 Armenian. |
| Tone of voice | The formal register (Դուք) 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 Armenian, 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. 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.
Three solutions meet every Standard requirement in Armenian, all three-step solutions (translation + LQA + repair). The dearest, on Google Gemini 3.1 Pro Preview at $597 per million characters, has the most total errors, 9.2 per 1,000 words, as many as the one on OpenAI GPT-5.5. The one on OpenAI GPT-5.5 takes the longest, 441 seconds per 1,000 words. The one on OpenAI GPT-5.6 Terra has the fewest errors, the lowest cost and the shortest time, so optimizing for quality, cost or latency selects it.
Every solution that meets even the Basic requirements in Armenian costs at least $556 per million characters and takes at least 254 seconds per 1,000 words. None of the solutions with one or two steps meet general translation quality. The closest are the two-step solutions (translation + post-edit) on OpenAI GPT-5.6 Terra and OpenAI GPT-5.5, with 21.8 and 20.1 weighted errors per 1,000 words. They cost $11.84 and $42.16 per million characters and take 152 and 331 seconds per 1,000 words.
The Advanced requirements add gender neutrality, and in Armenian the gender-neutrality agent is not needed to meet it. Translations from the three-step solutions (translation + LQA + repair) on OpenAI GPT-5.6 Terra and Google Gemini 3.1 Pro Preview have 100% of the sentences about people gender-neutral without it. So with these two solutions, meeting the Advanced requirements costs the same $556 and $597 per million characters, and takes the same 254 and 357 seconds per 1,000 words, as meeting the Standard ones.
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
Optimize for selects one of the solutions that meet the level: the one with the fewest total errors, the lowest cost or the shortest time. Where no solution meets the level, the card shows the available solution with the fewest total errors and what it misses. Total errors count every requirement with countable errors.
| Quality level | Tier | Solutions | Simplest solution | Model | Total errors per 1,000 words | Cost per million characters | Seconds per 1,000 words |
|---|---|---|---|---|---|---|---|
| Basic | Tier 2: two or more solutions, none simple | ☆☆☆ | three-step | OpenAI GPT-5.6 Terra | 8.6 | $556 | 254 |
| Standard | Tier 2: two or more solutions, none simple | ☆☆☆ | three-step | OpenAI GPT-5.6 Terra | 8.6 | $556 | 254 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | three-step | OpenAI GPT-5.6 Terra | 8.6 | $556 | 254 |
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 Armenian, the Standard requirements add tone of voice, tag handling and formatting to general translation quality. The three solutions that meet every Basic requirement, all three-step solutions (translation + LQA + repair), meet these too. The Advanced requirements add gender neutrality and leave two of them, on OpenAI GPT-5.6 Terra and Google Gemini 3.1 Pro Preview, as only the two best solutions count for gender neutrality. For all three sets, optimizing for quality, cost or latency selects OpenAI GPT-5.6 Terra.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| DeepSeek V3.2 | 1.35 | 67.9 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.2 | 65.3 | Misses a Standard requirement |
| z.ai GLM-5 | 4.49 | 55 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 4.67 | 26.7 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 5.26 | 42.5 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 7.07 | 31.3 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.7 | 28 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 18.89 | 29.3 | Misses a Standard requirement |
| Google Translation LLM | 21.72 | 76.1 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 68.6 | Misses a Standard requirement |
Among one-step solutions, the one on OpenAI GPT-5.6 Terra has the fewest weighted errors, 26.7 per 1,000 words, at $4.67 per million characters. The three cheaper ones, on DeepSeek V3.2, OpenAI GPT-4.1 and z.ai GLM-5, cost $1.35 to $4.49 per million characters and have 55.0 to 67.9 weighted errors per 1,000 words. The dearest with a conclusive Intento LQA result, on the LLM-based MT engines Google Translation LLM and DeepL Next Gen, cost $21.72 and $27.50 per million characters and have 76.1 and 68.6 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 48.8 |
| Tag handling | 0.3 |
| Tone of voice | 0.1 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
One-step solutions with a conclusive Intento LQA result have 26.7 to 76.1 weighted errors per 1,000 words, too many for general translation quality. Two-step solutions (translation + post-edit) on OpenAI GPT-5.5 and OpenAI GPT-5.6 Terra have 20.1 and 21.8 weighted errors per 1,000 words. Three-step solutions (translation + LQA + repair) on these two and Google Gemini 3.1 Pro Preview have 8.6 to 9.2 weighted errors per 1,000 words. One-step translations from Anthropic Claude Opus 4.8, OpenAI GPT-4.1 and DeepSeek V3.2 keep 99.2% to 99.6% of the tags.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Armenian | Class |
|---|---|---|
| Google Translation LLM | 76.1 | LLM-based MT engines |
| DeepL Next Gen | 68.6 | LLM-based MT engines |
| DeepSeek V3.2 | 67.9 | LLMs |
| OpenAI GPT-4.1 | 65.3 | LLMs |
| z.ai GLM-5 | 55 | LLMs |
| Google Gemini 3.5 Flash | 42.5 | LLMs |
| Google Gemini 3.1 Pro Preview | 31.3 | LLMs |
| Anthropic Claude Opus 4.8 | 29.3 | LLMs |
| OpenAI GPT-5.5 | 28 | LLMs |
| OpenAI GPT-5.6 Terra | 26.7 | LLMs |
No translation into Armenian: AI21 Jamba Large 1.7 (garbage or near-empty, cut off part-way), Swiss AI Apertus-8B-2509 (wrong language, wrong script, garbage or near-empty).
Eight of the 11 shortlisted models in Armenian are LLMs and three are LLM-based MT engines. In one step, translations from the LLMs have 26.7 to 67.9 weighted errors per 1,000 words, and those from DeepL Next Gen and Google Translation LLM have 68.6 and 76.1. The two LLMs outside our language-support criteria, OpenAI GPT-4.1 and DeepSeek V3.2, are those at 65.3 and 67.9 weighted errors per 1,000 words, over the 60 the criteria allow. None of the seven other MT engines we got a translation from made the shortlist.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| xAI Grok 4.20 (reasoning) | 0.801 | 9 | 1.79 | 16 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.801 | 10 | 1.69 | 12 | No | Not shortlisted |
| Google Gemma 4 | 0.799 | 13 | 1.61 | 11 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.794 | 14 | 1.59 | 10 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.793 | 15 | 1.70 | 14 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.792 | 16 | 1.87 | 18 | No | Not shortlisted |
| Mistral Large 3 | 0.789 | 17 | 1.69 | 13 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.783 | 19 | 2.11 | 22 | No | Not shortlisted |
| Microsoft Translator | 0.778 | 20 | 1.82 | 17 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.774 | 21 | 2.36 | 24 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.771 | 22 | 2.05 | 21 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.771 | 23 | 2.01 | 19 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.762 | 24 | 2.14 | 23 | No | Not shortlisted |
| Amazon Translate | 0.754 | 25 | 2.85 | 27 | No | Not shortlisted |
| MiniMax M2.5 | 0.749 | 26 | 2.75 | 26 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.727 | 27 | 2.67 | 25 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.693 | 28 | 4.00 | 28 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.636 | 29 | 5.68 | 29 | No | Not shortlisted |
| Cohere Command A Translate | 0.531 | 30 | 8.14 | 30 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.396 | 31 | 11.60 | 31 | No | Not shortlisted |
| OpenAI GPT-5.6 Terra | 0.812 | 1 | 1.31 | 1 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.810 | 2 | 1.39 | 2 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.807 | 3 | 1.48 | 5 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.804 | 4 | 1.52 | 8 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.803 | 5 | 1.40 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.802 | 6 | 1.49 | 7 | Yes | Top 8 by both metrics |
| z.ai GLM-5 | 0.801 | 7 | 1.49 | 6 | Yes | Top 8 by both metrics |
| DeepSeek V3.2 | 0.801 | 8 | 1.52 | 9 | Yes | Top 8 by COMET-Kiwi |
| Translated Lara 2 | 0.801 | 11 | 1.73 | 15 | Yes | Real-time place |
| OpenAI GPT-4.1 | 0.799 | 12 | 1.41 | 4 | Yes | Top 8 by MetricX |
| Google Translation LLM | 0.786 | 18 | 2.03 | 20 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| OpenAI GPT-5.6 Terra | Adjusted LLM | 1 | 0.812 | 1 | 1.31 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 2 | 0.810 | 2 | 1.39 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 3 | 0.807 | 5 | 1.48 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 4 | 0.804 | 8 | 1.52 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 5 | 0.803 | 3 | 1.40 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 6 | 0.802 | 7 | 1.49 | Top 8 by both metrics |
| z.ai GLM-5 | Adjusted LLM | 7 | 0.801 | 6 | 1.49 | Top 8 by both metrics |
| DeepSeek V3.2 | Adjusted LLM | 8 | 0.801 | 9 | 1.52 | Top 8 by COMET-Kiwi |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 9 | 0.801 | 16 | 1.79 | Not shortlisted |
| Anthropic Claude Fable 5 | Adjusted LLM | 10 | 0.801 | 12 | 1.69 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 11 | 0.801 | 15 | 1.73 | Real-time place |
| OpenAI GPT-4.1 | Adjusted LLM | 12 | 0.799 | 4 | 1.41 | Top 8 by MetricX |
| Google Gemma 4 | Adjusted LLM | 13 | 0.799 | 11 | 1.61 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.794 | 10 | 1.59 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 15 | 0.793 | 14 | 1.70 | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 16 | 0.792 | 18 | 1.87 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 17 | 0.789 | 13 | 1.69 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 18 | 0.786 | 20 | 2.03 | Real-time place |
| Mistral Medium 3.1 | Adjusted LLM | 19 | 0.783 | 22 | 2.11 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 20 | 0.778 | 17 | 1.82 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 21 | 0.774 | 24 | 2.36 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 22 | 0.771 | 21 | 2.05 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 23 | 0.771 | 19 | 2.01 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 24 | 0.762 | 23 | 2.14 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 25 | 0.754 | 27 | 2.85 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 26 | 0.749 | 26 | 2.75 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 27 | 0.727 | 25 | 2.67 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 28 | 0.693 | 28 | 4.00 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 29 | 0.636 | 29 | 5.68 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 30 | 0.531 | 30 | 8.14 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 31 | 0.396 | 31 | 11.60 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.809 | — | 1.43 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.807 | — | 1.50 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.795 | — | 1.88 | 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 31 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.
OpenAI GPT-4.1 is on the shortlist through MetricX alone, fourth there and 12th on COMET-Kiwi, and DeepSeek V3.2 through COMET-Kiwi alone, eighth there and ninth on MetricX. Both metrics rank Google Gemini 3.1 Pro Preview second, and Anthropic Claude Opus 4.8 sixth on COMET-Kiwi and seventh on MetricX. By Intento LQA, one-step translations from Anthropic Claude Opus 4.8 have fewer weighted errors, 29.3 per 1,000 words against 31.3. Nearest the cut in the tail is xAI Grok 4.20 (reasoning), ninth on COMET-Kiwi and 16th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, three-step | $556 |
| Standard | OpenAI GPT-5.6 Terra, three-step | $556 |
| Advanced | OpenAI GPT-5.6 Terra, three-step | $556 |
| Model | US dollars per million characters | Class |
|---|---|---|
| DeepSeek V3.2 | $1.35 | LLMs |
| OpenAI GPT-4.1 | $3.2 | LLMs |
| z.ai GLM-5 | $4.49 | LLMs |
| OpenAI GPT-5.6 Terra | $4.67 | LLMs |
| Google Gemini 3.5 Flash | $5.26 | LLMs |
| Google Gemini 3.1 Pro Preview | $7.07 | LLMs |
| OpenAI GPT-5.5 | $11.7 | LLMs |
| Anthropic Claude Opus 4.8 | $18.89 | LLMs |
| Google Translation LLM | $21.72 | LLM-based MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Model | Tokens per character | Class |
|---|---|---|
| MiniMax M2.5 | 1.152 | LLMs |
| Cohere Command A Translate | 0.94 | LLMs |
| Mistral Large 3 | 0.932 | LLMs |
| Mistral Medium 3.1 | 0.922 | LLMs |
| z.ai GLM-5 | 0.835 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.834 | Open-weight LLMs |
| Anthropic Claude Fable 5 | 0.602 | LLMs |
| Anthropic Claude Opus 4.8 | 0.599 | LLMs |
| Anthropic Claude Sonnet 5 | 0.597 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.572 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.565 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.502 | LLMs |
| MoonshotAI Kimi K2.5 | 0.502 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.501 | LLMs |
| DeepSeek V3.2 | 0.495 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.475 | LLMs |
| Google Gemini 3.5 Flash | 0.474 | LLMs |
| Google Gemma 4 | 0.473 | Open-weight LLMs |
| OpenAI GPT-4.1 | 0.307 | LLMs |
| OpenAI GPT-5.5 | 0.307 | LLMs |
| OpenAI GPT-5.6 Sol | 0.307 | LLMs |
| OpenAI GPT-5.6 Terra | 0.307 | LLMs |
| OpenAI GPT-5.6 Luna | 0.301 | LLMs |
The tokenizer of z.ai GLM-5 produces 0.835 tokens per character of the Armenian translation, against 0.307 for OpenAI GPT-5.6 Terra. Even so, the one-step solution on z.ai GLM-5 costs $4.49 per million characters, less than the $4.67 of the one on OpenAI GPT-5.6 Terra. The steps change the cost far more. On Google Gemini 3.1 Pro Preview, the one-step solution costs $7.07 per million characters, the two-step solution (translation + post-edit) $40.70 and the three-step solution (translation + LQA + repair) $597.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, three-step | 254 |
| Standard | OpenAI GPT-5.6 Terra, three-step | 254 |
| Advanced | OpenAI GPT-5.6 Terra, three-step | 254 |
| 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 | 0.4 | MT engines |
| Google Translation LLM | 0.6 | LLM-based MT engines |
| DeepL Next Gen | 0.7 | LLM-based MT engines |
| Amazon Translate | 1.2 | MT engines |
| Translated Lara 2 | 1.6 | LLM-based MT engines |
| SYSTRAN PNMT API | 2.4 | MT engines |
| OpenAI gpt-oss-120b | 2.7 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 5.8 | MT engines |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| OpenAI GPT-4.1 | 24 | LLMs |
| Alibaba Machine Translation API | 31 | MT engines |
| xAI Grok 4.20 (non-reasoning) | 35 | LLMs |
| Mistral Medium 3.1 | 52 | LLMs |
| OpenAI GPT-5.6 Terra | 63 | LLMs |
| DeepSeek V3.2 | 65 | LLMs |
| Anthropic Claude Haiku 4.5 | 73 | LLMs |
| Mistral Large 3 | 77 | LLMs |
| xAI Grok 4.20 (reasoning) | 81 | LLMs |
| Google Gemini 3.5 Flash | 88 | LLMs |
| Cohere Command A Translate | 116 | LLMs |
| MiniMax M2.5 | 118 | LLMs |
| Google Gemini 3.1 Pro Preview | 120 | LLMs |
| OpenAI GPT-5.6 Sol | 156 | LLMs |
| OpenAI GPT-5.5 | 159 | LLMs |
| Anthropic Claude Sonnet 4.6 | 174 | LLMs |
| Anthropic Claude Sonnet 5 | 202 | LLMs |
| Anthropic Claude Fable 5 | 222 | LLMs |
| Anthropic Claude Opus 4.8 | 243 | LLMs |
| MoonshotAI Kimi K2.5 | 249 | LLMs |
| Google Gemma 4 | 255 | Open-weight LLMs |
| Alibaba Qwen3-235B-A22B | 309 | Open-weight LLMs |
| z.ai GLM-5 | 378 | 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 Armenian.
The nine real-time models in Armenian are five MT engines, three LLM-based MT engines and the open-weight LLM OpenAI gpt-oss-120b. The three on the shortlist are the LLM-based MT engines. One-step translations from DeepL Next Gen and Google Translation LLM have 68.6 and 76.1 weighted errors per 1,000 words, and the solution on Translated Lara 2 has no conclusive Intento LQA result. The fastest solution that meets every Basic requirement takes 254 seconds per 1,000 words, and so does the fastest for the Advanced ones.
| 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 Terra | three-step | 100% | ✓ | $556 | 254 |
| Google Gemini 3.1 Pro Preview | three-step | 100% | ✓ | $597 | 357 |
In Armenian, the low ends of the agreement of the quality-estimation pair and the similarity pair, 0.66 and 0.48, are above their thresholds of 0.53 and 0.46, so neither result is borderline.
Leave feedback