The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Farsi 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 Farsi. |
| Tone of voice | The formal register (شما) 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 Farsi, with a column for each requirement you select. Under Optimize for, choose which of them comes first: the one with the fewest total errors, the lowest cost or the shortest time. Click a column heading to sort by it.
The shortlisted solutions on 12 models, with their Intento LQA results. Total errors per 1,000 words add the weighted general translation quality errors (minor 1, major 5, critical 10) to the missed glossary terms, sentences in the wrong register, tag errors and sentences that fail gender neutrality, each counted once. Formatting counts as met or missed and does not add to the total errors. The models that did not make the shortlist have automatic scores only: the shortlist. Gender neutrality: the Advanced requirements count the two best solutions of the language, with the gender-neutrality agent where they need it. The results of the other solutions are shown for information.
Under the Standard requirements in Farsi, a cost cap below $10.99 per million characters or a time limit of at most 77 seconds per 1,000 words leaves only the one-step solution on Google Gemini 3.1 Pro Preview of the three solutions that meet them. A cap below $4.41 per million characters or a limit below 68 seconds per 1,000 words removes all three.
Meeting the Advanced requirements in Farsi costs and takes the same as meeting the Standard ones. The two best solutions meet gender neutrality without the gender-neutrality agent. So the one-step solution on Google Gemini 3.1 Pro Preview stays at $4.41 per million characters and 68 seconds per 1,000 words, and the two-step solution (translation + post-edit) on OpenAI GPT-5.5 at $10.99 per million characters and 81 seconds per 1,000 words.
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
one-step LLM adjusted for Standard requirements
Google Gemini 3.1 Pro Preview
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 | Google Gemini 3.1 Pro Preview | 12.9 | $4.41 | 68 |
| Standard | Tier 1: simple solution + backup | ★ ☆☆ | one-step | Google Gemini 3.1 Pro Preview | 12.9 | $4.41 | 68 |
| Advanced | Tier 1: simple solution + backup | ★ ☆ | one-step | Google Gemini 3.1 Pro Preview | 12.9 | $4.41 | 68 |
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.
Tag handling swaps the one-step solutions on OpenAI GPT-5.5 and Anthropic Claude Fable 5 for their two-step solutions (translation + post-edit) under the Standard requirements. The Advanced requirements keep the two best solutions and drop the two-step solution on Anthropic Claude Fable 5. Optimize for selects Google Gemini 3.1 Pro Preview for quality, cost and latency at every level.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.13 | 32 | Misses a Standard requirement |
| OpenAI GPT-5.6 Luna | 0.44 | 22.1 | Misses a Standard requirement |
| z.ai GLM-5 | 2.05 | 30.3 | Misses a Standard requirement |
| OpenAI GPT-4.1 | 3.03 | 40.9 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.27 | 39.2 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 4.41 | 12.9 | Meets every Standard requirement |
| Anthropic Claude Sonnet 4.6 | 9.62 | 36.9 | Misses a Standard requirement |
| Microsoft Translator | 10 | 118.7 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 10.99 | 15.2 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 34.6 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 38.2 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 34.32 | 17.5 | Misses a Standard requirement |
The cost and errors front is a single LLM solution. Of the one-step solutions, the ones on the LLM-based MT engines DeepL Next Gen and Translated Lara 2 cost $27.50 and $24.99 per million characters and have 38.2 and 34.6 weighted errors per 1,000 words. The one on the MT engine Microsoft Translator costs $10.00 per million characters and has 118.7 weighted errors per 1,000 words. The two cheapest, on Google Gemma 4 and OpenAI GPT-5.6 Luna at $0.13 and $0.44 per million characters, have 32.0 and 22.1 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 33.3 |
| Tag handling | 1 |
| Tone of voice | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
General translation quality and tag handling hold the one-step solutions back, and only the one on Google Gemini 3.1 Pro Preview meets both. Those on OpenAI GPT-5.6 Luna and z.ai GLM-5 meet tag handling, with 22.1 and 30.3 weighted errors per 1,000 words, over the bar of general translation quality. Their two-step solutions (translation + post-edit) have 23.7 and 31.3 weighted errors per 1,000 words. None of the solutions on Microsoft Translator, DeepL Next Gen and Translated Lara 2 meet formatting, and the nine on LLMs all do.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Farsi | Class |
|---|---|---|
| Microsoft Translator | 118.7 | MT engines |
| OpenAI GPT-4.1 | 40.9 | LLMs |
| Google Gemini 3.5 Flash | 39.2 | LLMs |
| DeepL Next Gen | 38.2 | LLM-based MT engines |
| Anthropic Claude Sonnet 4.6 | 36.9 | LLMs |
| Translated Lara 2 | 34.6 | LLM-based MT engines |
| Google Gemma 4 | 32 | Open-weight LLMs |
| z.ai GLM-5 | 30.3 | LLMs |
| OpenAI GPT-5.6 Luna | 22.1 | LLMs |
| Anthropic Claude Fable 5 | 17.5 | LLMs |
| OpenAI GPT-5.5 | 15.2 | LLMs |
| Google Gemini 3.1 Pro Preview | 12.9 | LLMs |
The three one-step solutions that meet general translation quality in Farsi all run on LLMs. The eight LLM solutions on the shortlist have 12.9 to 40.9 weighted errors per 1,000 words, and the one on the open-weight Google Gemma 4 has 32.0. The three solutions on MT engines and LLM-based MT engines have 34.6 to 118.7 weighted errors per 1,000 words. All nine shortlisted LLMs support Farsi by our measured criteria.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Cohere Command A Translate | 0.809 | 10 | 1.08 | 12 | No | Not shortlisted |
| Anthropic Claude Opus 4.8 | 0.807 | 12 | 1.14 | 16 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.807 | 13 | 1.04 | 10 | No | Not shortlisted |
| Mistral Large 3 | 0.807 | 14 | 1.09 | 13 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.804 | 16 | 1.15 | 18 | No | Not shortlisted |
| DeepSeek V3.2 | 0.804 | 17 | 1.13 | 15 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.802 | 19 | 1.17 | 19 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.798 | 20 | 1.21 | 21 | No | Not shortlisted |
| MiniMax M2.5 | 0.798 | 21 | 1.17 | 20 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.796 | 22 | 1.14 | 17 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.795 | 23 | 1.28 | 23 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.793 | 24 | 1.44 | 25 | No | Not shortlisted |
| Google Translation LLM | 0.793 | 25 | 1.63 | 26 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.783 | 26 | 1.33 | 24 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.782 | 27 | 1.94 | 29 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.751 | 28 | 1.87 | 28 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.750 | 29 | 1.85 | 27 | No | Not shortlisted |
| Amazon Translate | 0.740 | 30 | 1.98 | 30 | No | Not shortlisted |
| Cohere Tiny Aya | 0.732 | 31 | 2.41 | 32 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.728 | 32 | 2.30 | 31 | No | Not shortlisted |
| AppTek NMT | 0.668 | 33 | 3.84 | 33 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.567 | 34 | 5.27 | 34 | No | Not shortlisted |
| Google Gemini 3.1 Pro Preview | 0.816 | 1 | 0.94 | 2 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.814 | 2 | 1.02 | 7 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.814 | 3 | 0.99 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.813 | 4 | 1.02 | 9 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | 0.813 | 5 | 1.02 | 8 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | 0.812 | 6 | 0.93 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-4.1 | 0.811 | 7 | 0.97 | 3 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.810 | 8 | 1.07 | 11 | Yes | Top 8 by COMET-Kiwi |
| Google Gemma 4 | 0.810 | 9 | 0.99 | 5 | Yes | Top 8 by MetricX |
| z.ai GLM-5 | 0.807 | 11 | 1.01 | 6 | Yes | Top 8 by MetricX |
| Translated Lara 2 | 0.805 | 15 | 1.26 | 22 | Yes | Real-time place |
| Microsoft Translator | 0.803 | 18 | 1.12 | 14 | 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 | 0.816 | 2 | 0.94 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 2 | 0.814 | 7 | 1.02 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 3 | 0.814 | 4 | 0.99 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 4 | 0.813 | 9 | 1.02 | Top 8 by COMET-Kiwi |
| Google Gemini 3.5 Flash | Adjusted LLM | 5 | 0.813 | 8 | 1.02 | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 6 | 0.812 | 1 | 0.93 | Top 8 by both metrics |
| OpenAI GPT-4.1 | Adjusted LLM | 7 | 0.811 | 3 | 0.97 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 8 | 0.810 | 11 | 1.07 | Top 8 by COMET-Kiwi |
| Google Gemma 4 | Adjusted LLM | 9 | 0.810 | 5 | 0.99 | Top 8 by MetricX |
| Cohere Command A Translate | Adjusted LLM | 10 | 0.809 | 12 | 1.08 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 11 | 0.807 | 6 | 1.01 | Top 8 by MetricX |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 12 | 0.807 | 16 | 1.14 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 13 | 0.807 | 10 | 1.04 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 14 | 0.807 | 13 | 1.09 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 15 | 0.805 | 22 | 1.26 | Real-time place |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 16 | 0.804 | 18 | 1.15 | Not shortlisted |
| DeepSeek V3.2 | Adjusted LLM | 17 | 0.804 | 15 | 1.13 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 18 | 0.803 | 14 | 1.12 | Real-time place |
| Mistral Medium 3.1 | Adjusted LLM | 19 | 0.802 | 19 | 1.17 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 20 | 0.798 | 21 | 1.21 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 21 | 0.798 | 20 | 1.17 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 22 | 0.796 | 17 | 1.14 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 23 | 0.795 | 23 | 1.28 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 24 | 0.793 | 25 | 1.44 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 25 | 0.793 | 26 | 1.63 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 26 | 0.783 | 24 | 1.33 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 27 | 0.782 | 29 | 1.94 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 28 | 0.751 | 28 | 1.87 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 29 | 0.750 | 27 | 1.85 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 30 | 0.740 | 30 | 1.98 | Not shortlisted |
| Cohere Tiny Aya | Adjusted LLM | 31 | 0.732 | 32 | 2.41 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 32 | 0.728 | 31 | 2.30 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 33 | 0.668 | 33 | 3.84 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 34 | 0.567 | 34 | 5.27 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.813 | — | 1.00 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.809 | — | 1.06 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.796 | — | 1.16 | 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 34 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-5.5 is 8th on COMET-Kiwi and 11th on MetricX, so COMET-Kiwi alone put it on the shortlist. By Intento LQA, its translations have 15.2 weighted errors per 1,000 words, the fewest of the one-step solutions after Google Gemini 3.1 Pro Preview. MetricX alone put Google Gemma 4 and z.ai GLM-5 on the list, 5th and 6th there, and neither of their solutions meets general translation quality. The tail models nearest the cut are Cohere Command A Translate, 10th on COMET-Kiwi, and Anthropic Claude Sonnet 5, 10th on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $4.41 |
| Standard | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $4.41 |
| Advanced | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | $4.41 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.13 | Open-weight LLMs |
| OpenAI GPT-5.6 Luna | $0.44 | LLMs |
| z.ai GLM-5 | $2.05 | LLMs |
| OpenAI GPT-4.1 | $3.03 | LLMs |
| Google Gemini 3.5 Flash | $3.27 | LLMs |
| Google Gemini 3.1 Pro Preview | $4.41 | LLMs |
| Anthropic Claude Sonnet 4.6 | $9.62 | LLMs |
| Microsoft Translator | $10 | MT engines |
| OpenAI GPT-5.5 | $10.99 | LLMs |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $34.32 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Mistral Large 3 | 0.833 | LLMs |
| Mistral Medium 3.1 | 0.833 | LLMs |
| Anthropic Claude Sonnet 5 | 0.627 | LLMs |
| Anthropic Claude Opus 4.8 | 0.626 | LLMs |
| Anthropic Claude Fable 5 | 0.624 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.597 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.589 | LLMs |
| AI21 Jamba Large 1.7 | 0.583 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.526 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 0.437 | LLMs |
| MiniMax M2.5 | 0.427 | LLMs |
| z.ai GLM-5 | 0.385 | LLMs |
| DeepSeek V3.2 | 0.381 | LLMs |
| Cohere Command A Translate | 0.34 | LLMs |
| OpenAI GPT-5.6 Terra | 0.332 | LLMs |
| OpenAI GPT-5.6 Sol | 0.331 | LLMs |
| OpenAI GPT-5.6 Luna | 0.329 | LLMs |
| OpenAI GPT-4.1 | 0.328 | LLMs |
| OpenAI GPT-5.5 | 0.328 | LLMs |
| Google Gemini 3.5 Flash | 0.322 | LLMs |
| Google Gemma 4 | 0.322 | Open-weight LLMs |
| Google Gemini 3.1 Pro Preview | 0.321 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.308 | LLMs |
| Cohere Tiny Aya | 0.306 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.306 | LLMs |
With billing on tokens, the tokenizers produce a median 0.38 tokens per character of Farsi, 1.6 times their count for English. Among the shortlisted LLMs, the tokenizers of Anthropic Claude Fable 5 and Anthropic Claude Sonnet 4.6 produce the most, 0.624 and 0.597 tokens per character, and the tokenizer of Google Gemini 3.1 Pro Preview the fewest, 0.321. The cheapest solution is the same at all three levels, the one-step solution on Google Gemini 3.1 Pro Preview at $4.41 per million characters.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | 68 |
| Standard | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | 68 |
| Advanced | Google Gemini 3.1 Pro Preview, one-step LLM adjusted for Standard requirements | 68 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.3 | MT engines |
| Google Translation LLM | 0.5 | LLM-based MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Microsoft Translator | 0.6 | MT engines |
| Amazon Translate | 1.3 | MT engines |
| Translated Lara 2 | 1.4 | LLM-based MT engines |
| SYSTRAN PNMT API | 1.4 | MT engines |
| OpenAI gpt-oss-120b | 2.3 | Open-weight LLMs |
| Yandex Cloud Translate v2 | 2.4 | MT engines |
| xAI Grok 4.20 (non-reasoning) | 14 | LLMs |
| Cohere Tiny Aya | 15 | LLMs |
| OpenAI GPT-5.6 Luna | 16 | LLMs |
| Google Gemini 3.5 Flash | 17 | LLMs |
| Alibaba Machine Translation API | 24 | MT engines |
| OpenAI GPT-4.1 | 26 | LLMs |
| OpenAI GPT-5.6 Terra | 32 | LLMs |
| DeepSeek V3.2 | 38 | LLMs |
| Cohere Command A Translate | 39 | LLMs |
| MiniMax M2.5 | 41 | LLMs |
| Swiss AI Apertus-8B-2509 | 46 | Open-weight LLMs |
| xAI Grok 4.20 (reasoning) | 48 | LLMs |
| Anthropic Claude Haiku 4.5 | 51 | LLMs |
| Mistral Medium 3.1 | 52 | LLMs |
| Mistral Large 3 | 59 | LLMs |
| Google Gemini 3.1 Pro Preview | 68 | LLMs |
| Anthropic Claude Sonnet 5 | 72 | LLMs |
| AI21 Jamba Large 1.7 | 76 | LLMs |
| Anthropic Claude Fable 5 | 77 | LLMs |
| AppTek NMT | 79 | MT engines |
| OpenAI GPT-5.5 | 81 | LLMs |
| OpenAI GPT-5.6 Sol | 87 | LLMs |
| Anthropic Claude Opus 4.8 | 94 | LLMs |
| Anthropic Claude Sonnet 4.6 | 101 | LLMs |
| Google Gemma 4 | 144 | Open-weight LLMs |
| Alibaba Qwen3-235B-A22B | 175 | Open-weight LLMs |
| z.ai GLM-5 | 194 | LLMs |
| MoonshotAI Kimi K2.5 | 307 | 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 Farsi.
Seven of the eight real-time models in Farsi are MT engines or LLM-based MT engines, and the other one is the open-weight OpenAI gpt-oss-120b, which did not make the shortlist. None of the three real-time solutions on the shortlist, on DeepL Next Gen, Translated Lara 2 and Microsoft Translator, meet general translation quality, tag handling or formatting. The fastest solution is the same at all three levels, the one-step solution on Google Gemini 3.1 Pro Preview at 68 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 |
|---|---|---|---|---|---|
| Google Gemini 3.1 Pro Preview | one-step | 100% | ✓ | $4.41 | 68 |
| OpenAI GPT-5.5 | two-step | 100% | ✓ | $10.99 | 81 |
Tone of voice in Farsi is the formal register, and translations from all 12 shortlisted one-step solutions have 100% of their sentences in it. For gender neutrality, a translation must not add a gender the source does not specify, and we never allow a default gender. Farsi has no grammatical gender for persons, so a translation is neutral by default, and gender neutrality here means avoiding gendered terms. The two best solutions have 100% of the sentences about people gender-neutral.
The agreement of both metric pairs in Farsi is clear of the borderline, and the three LQA judges agree on a median of 96% of the sentences of the 12 shortlisted solutions. So the Intento LQA results on this page rest on measurement we can rely on.
Leave feedback