The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Georgian 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 Georgian. |
| Tone of voice | The formal register (თქვენ) 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 Georgian, 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.
Two solutions meet every Standard requirement in Georgian. The one-step solution on Anthropic Claude Fable 5 costs $39.95 per million characters and takes 259 seconds per 1,000 words, and its translations have 16.2 total errors per 1,000 words. The two-step solution (translation + post-edit) on OpenAI GPT-5.5 costs $11.71 per million characters and takes 120 seconds per 1,000 words, with 17.8 total errors per 1,000 words. So Optimize for selects Anthropic Claude Fable 5 for quality and OpenAI GPT-5.5 for cost and latency.
A limit of one step leaves only the solution on Anthropic Claude Fable 5 among those that meet every Standard requirement. A cost cap of at most $39.94 per million characters or a time limit below 259 seconds per 1,000 words leaves only the two-step solution on OpenAI GPT-5.5. None of the solutions that meet every Standard requirement fit a cost cap of at most $11.70 per million characters or a time limit below 120 seconds per 1,000 words.
The Advanced requirements add gender neutrality to the Standard ones. Translations from both solutions that meet the Standard requirements have 100% of the sentences about people gender-neutral without the gender-neutrality agent. So meeting every Advanced requirement costs the same $11.71 or $39.95 per million characters and takes the same 120 or 259 seconds per 1,000 words, and Optimize for selects the same solutions.
Anthropic Claude Fable 5
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
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
two-step
OpenAI GPT-5.5
two-step
Anthropic Claude Fable 5
one-step LLM adjusted for Standard requirements
OpenAI GPT-5.5
two-step
OpenAI GPT-5.5
two-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 1: simple solution + backup | ★★ | one-step | OpenAI GPT-5.5 | 17.8 | $11.71 | 120 |
| Standard | Tier 1: simple solution + backup | ★ ☆ | one-step | Anthropic Claude Fable 5 | 16.2 | $39.95 | 259 |
| Advanced | Tier 1: simple solution + backup | ★ ☆ | one-step | Anthropic Claude Fable 5 | 16.2 | $39.95 | 259 |
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.
Under the Basic requirements, both solutions take one step, and Optimize for selects Anthropic Claude Fable 5 for quality and OpenAI GPT-5.5 for cost and latency. Under the Standard requirements, the one-step solution on OpenAI GPT-5.5 misses formatting. The two-step solution (translation + post-edit) on the same model meets it and takes its place for cost and latency. Both solutions also meet every Advanced requirement without the gender-neutrality agent, so Optimize for selects the same ones.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| OpenAI GPT-5.6 Luna | 0.47 | 23.4 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 4.46 | 40.5 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 6.01 | 36.3 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 11.11 | 44.2 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.71 | 17.8 | Misses a Standard requirement |
| Anthropic Claude Opus 4.8 | 20 | 33.9 | Misses a Standard requirement |
| Google Translation LLM | 21.74 | 80.1 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 51.1 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 74.5 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 39.95 | 16.2 | Meets every Standard requirement |
Among the one-step solutions in Georgian, a higher price does not always come with fewer errors. The solution on Anthropic Claude Opus 4.8 costs $20 per million characters, and its translations have 33.9 weighted errors per 1,000 words, against $11.71 and 17.8 for OpenAI GPT-5.5. The solutions on the three LLM-based MT engines cost $21.74 to $27.50 per million characters, and their translations have 51.1 to 80.1 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 38.4 |
| Tag handling | 0.7 |
| Tone of voice | 0 |
| 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 Georgian | Class |
|---|---|---|
| Google Translation LLM | 80.1 | LLM-based MT engines |
| DeepL Next Gen | 74.5 | LLM-based MT engines |
| Translated Lara 2 | 51.1 | LLM-based MT engines |
| Anthropic Claude Sonnet 4.6 | 44.2 | LLMs |
| Google Gemini 3.5 Flash | 40.5 | LLMs |
| Google Gemini 3.1 Pro Preview | 36.3 | LLMs |
| Anthropic Claude Opus 4.8 | 33.9 | LLMs |
| OpenAI GPT-5.6 Luna | 23.4 | LLMs |
| OpenAI GPT-5.5 | 17.8 | LLMs |
| Anthropic Claude Fable 5 | 16.2 | LLMs |
No translation into Georgian: AI21 Jamba Large 1.7 (cut off part-way).
The Intento LQA results split by model type in Georgian. Translations from the seven shortlisted LLMs with a conclusive result have 16.2 to 44.2 weighted errors per 1,000 words, and translations from the three LLM-based MT engines 51.1 to 80.1. On the shortlist, one-step solutions on LLMs cost a median $8.56 per million characters, against $24.99 on LLM-based MT engines. The eighth shortlisted LLM, z.ai GLM-5, is the one solution without a conclusive Intento LQA result.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| DeepSeek V3.2 | 0.789 | 10 | 1.55 | 15 | No | Not shortlisted |
| xAI Grok 4.20 (non-reasoning) | 0.787 | 12 | 1.44 | 12 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.786 | 13 | 1.28 | 9 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.783 | 14 | 1.63 | 16 | No | Not shortlisted |
| Google Gemma 4 | 0.778 | 16 | 1.43 | 11 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.776 | 17 | 1.66 | 17 | No | Not shortlisted |
| Mistral Large 3 | 0.774 | 18 | 1.51 | 13 | No | Not shortlisted |
| MiniMax M2.5 | 0.759 | 19 | 2.02 | 21 | No | Not shortlisted |
| Microsoft Translator | 0.758 | 20 | 1.91 | 19 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.755 | 21 | 2.25 | 24 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.753 | 22 | 1.95 | 20 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.751 | 23 | 2.11 | 22 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.744 | 24 | 2.15 | 23 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.726 | 25 | 2.48 | 25 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.712 | 26 | 2.74 | 26 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.711 | 27 | 3.36 | 28 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.690 | 28 | 3.22 | 27 | No | Not shortlisted |
| Cohere Command A Translate | 0.680 | 29 | 4.36 | 29 | No | Not shortlisted |
| Amazon Translate | 0.612 | 30 | 4.70 | 30 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.605 | 31 | 4.77 | 31 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.234 | 32 | 15.49 | 32 | No | Not shortlisted |
| Anthropic Claude Fable 5 | 0.801 | 1 | 1.28 | 8 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.801 | 2 | 1.20 | 4 | Yes | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | 0.799 | 3 | 1.33 | 10 | Yes | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | 0.798 | 4 | 1.09 | 1 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | 0.797 | 5 | 1.14 | 2 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.797 | 6 | 1.15 | 3 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.793 | 7 | 1.24 | 6 | Yes | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | 0.793 | 8 | 1.26 | 7 | Yes | Top 8 by both metrics |
| Translated Lara 2 | 0.791 | 9 | 1.53 | 14 | Yes | Real-time place |
| z.ai GLM-5 | 0.789 | 11 | 1.22 | 5 | Yes | Top 8 by MetricX |
| Google Translation LLM | 0.779 | 15 | 1.69 | 18 | Yes | Real-time place |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Anthropic Claude Fable 5 | Adjusted LLM | 1 | 0.801 | 8 | 1.28 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 2 | 0.801 | 4 | 1.20 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 3 | 0.799 | 10 | 1.33 | Top 8 by COMET-Kiwi |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 4 | 0.798 | 1 | 1.09 | Top 8 by both metrics |
| OpenAI GPT-5.6 Luna | Adjusted LLM | 5 | 0.797 | 2 | 1.14 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 6 | 0.797 | 3 | 1.15 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 7 | 0.793 | 6 | 1.24 | Top 8 by both metrics |
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 8 | 0.793 | 7 | 1.26 | Top 8 by both metrics |
| Translated Lara 2 | Adjusted NMT | 9 | 0.791 | 14 | 1.53 | Real-time place |
| DeepSeek V3.2 | Adjusted LLM | 10 | 0.789 | 15 | 1.55 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 11 | 0.789 | 5 | 1.22 | Top 8 by MetricX |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | 12 | 0.787 | 12 | 1.44 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 13 | 0.786 | 9 | 1.28 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 14 | 0.783 | 16 | 1.63 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 15 | 0.779 | 18 | 1.69 | Real-time place |
| Google Gemma 4 | Adjusted LLM | 16 | 0.778 | 11 | 1.43 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 17 | 0.776 | 17 | 1.66 | Not shortlisted |
| Mistral Large 3 | Adjusted LLM | 18 | 0.774 | 13 | 1.51 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 19 | 0.759 | 21 | 2.02 | Not shortlisted |
| Microsoft Translator | Off-the-shelf NMT | 20 | 0.758 | 19 | 1.91 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 21 | 0.755 | 24 | 2.25 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 22 | 0.753 | 20 | 1.95 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Off-the-shelf LLM | 23 | 0.751 | 22 | 2.11 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 24 | 0.744 | 23 | 2.15 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 25 | 0.726 | 25 | 2.48 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 26 | 0.712 | 26 | 2.74 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 27 | 0.711 | 28 | 3.36 | Not shortlisted |
| OpenAI gpt-oss-120b | Adjusted LLM | 28 | 0.690 | 27 | 3.22 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 29 | 0.680 | 29 | 4.36 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 30 | 0.612 | 30 | 4.70 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 31 | 0.605 | 31 | 4.77 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 32 | 0.234 | 32 | 15.49 | Not shortlisted |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.791 | — | 1.10 | Same model family as OpenAI GPT-5.6 Luna |
| OpenAI GPT-5.6 Terra | Adjusted LLM | — | 0.797 | — | 1.18 | Same model family as OpenAI GPT-5.6 Luna |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | — | 0.783 | — | 1.68 | 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 32 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 in Georgian. Translations from Google Gemini 3.1 Pro Preview, at rank 1 on MetricX, have 36.3 weighted errors per 1,000 words, and those from Anthropic Claude Fable 5, at rank 8 on MetricX, have 16.2. COMET-Kiwi alone brought Anthropic Claude Opus 4.8 onto the list, at rank 3, and MetricX alone z.ai GLM-5, at rank 5. The tail models nearest the cut are DeepSeek V3.2, at rank 10 on COMET-Kiwi, and OpenAI GPT-4.1, at rank 9 on MetricX.
| Quality level | Cheapest solution that meets every requirement of the level | Cost per million characters |
|---|---|---|
| Basic | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | $11.71 |
| Standard | OpenAI GPT-5.5, two-step | $11.71 |
| Advanced | OpenAI GPT-5.5, two-step | $11.71 |
| Model | US dollars per million characters | Class |
|---|---|---|
| OpenAI GPT-5.6 Luna | $0.47 | LLMs |
| z.ai GLM-5 | $4.42 | LLMs |
| Google Gemini 3.5 Flash | $4.46 | LLMs |
| Google Gemini 3.1 Pro Preview | $6.01 | LLMs |
| Anthropic Claude Sonnet 4.6 | $11.11 | LLMs |
| OpenAI GPT-5.5 | $11.71 | LLMs |
| Anthropic Claude Opus 4.8 | $20 | LLMs |
| Google Translation LLM | $21.74 | LLM-based MT engines |
| Translated Lara 2 | $24.99 | LLM-based MT engines |
| DeepL Next Gen | $27.5 | LLM-based MT engines |
| Anthropic Claude Fable 5 | $39.95 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Cohere Command A Translate | 0.888 | LLMs |
| Mistral Large 3 | 0.86 | LLMs |
| Mistral Medium 3.1 | 0.849 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.836 | Open-weight LLMs |
| z.ai GLM-5 | 0.831 | LLMs |
| MiniMax M2.5 | 0.813 | LLMs |
| MoonshotAI Kimi K2.5 | 0.693 | LLMs |
| Anthropic Claude Opus 4.8 | 0.651 | LLMs |
| Anthropic Claude Fable 5 | 0.647 | LLMs |
| Anthropic Claude Sonnet 5 | 0.646 | LLMs |
| DeepSeek V3.2 | 0.63 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.621 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.618 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.559 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.558 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.402 | LLMs |
| Google Gemini 3.5 Flash | 0.401 | LLMs |
| Google Gemma 4 | 0.399 | Open-weight LLMs |
| OpenAI GPT-4.1 | 0.319 | LLMs |
| OpenAI GPT-5.6 Terra | 0.317 | LLMs |
| OpenAI GPT-5.6 Luna | 0.316 | LLMs |
| OpenAI GPT-5.5 | 0.315 | LLMs |
| OpenAI GPT-5.6 Sol | 0.315 | LLMs |
With billing on tokens, the tokenizers of the 23 models we measured produce a median 0.62 tokens per character of Georgian, 2.4 times their count for English. The tokenizer of Anthropic Claude Fable 5 produces 0.65 tokens per character, against 0.32 for OpenAI GPT-5.5. Meeting the Basic, Standard or Advanced requirements costs $11.71 per million characters at the least, with solutions on OpenAI GPT-5.5.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.5, one-step LLM adjusted for Standard requirements | 120 |
| Standard | OpenAI GPT-5.5, two-step | 120 |
| Advanced | OpenAI GPT-5.5, two-step | 120 |
| 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 | 0.4 | MT engines |
| Google Translation LLM | 0.7 | LLM-based MT engines |
| DeepL Next Gen | 0.8 | LLM-based MT engines |
| Amazon Translate | 1.6 | MT engines |
| Translated Lara 2 | 1.8 | LLM-based MT engines |
| Yandex Cloud Translate v2 | 2.4 | MT engines |
| OpenAI gpt-oss-120b | 2.6 | Open-weight LLMs |
| SYSTRAN PNMT API | 5.3 | MT engines |
| OpenAI GPT-5.6 Luna | 15 | LLMs |
| Google Gemini 3.5 Flash | 17 | LLMs |
| Alibaba Machine Translation API | 27 | MT engines |
| OpenAI GPT-4.1 | 34 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 35 | LLMs |
| Swiss AI Apertus-8B-2509 | 58 | Open-weight LLMs |
| Mistral Medium 3.1 | 59 | LLMs |
| OpenAI GPT-5.6 Terra | 63 | LLMs |
| Anthropic Claude Haiku 4.5 | 70 | LLMs |
| xAI Grok 4.20 (reasoning) | 80 | LLMs |
| Mistral Large 3 | 90 | LLMs |
| MiniMax M2.5 | 101 | LLMs |
| DeepSeek V3.2 | 102 | LLMs |
| Google Gemini 3.1 Pro Preview | 104 | LLMs |
| OpenAI GPT-5.6 Sol | 108 | LLMs |
| Cohere Command A Translate | 117 | LLMs |
| OpenAI GPT-5.5 | 120 | LLMs |
| Anthropic Claude Sonnet 4.6 | 125 | LLMs |
| Google Gemma 4 | 163 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 199 | LLMs |
| Anthropic Claude Sonnet 5 | 199 | LLMs |
| Anthropic Claude Fable 5 | 259 | LLMs |
| Anthropic Claude Opus 4.8 | 270 | LLMs |
| Alibaba Qwen3-235B-A22B | 294 | Open-weight LLMs |
| z.ai GLM-5 | 342 | 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 Georgian.
All real-time models (<2 sec) in Georgian but the open-weight LLM OpenAI gpt-oss-120b are MT engines or LLM-based MT engines. Translations from the three of them on the shortlist have 10 to 21 major errors, more than general translation quality allows. The shortlisted LLM solutions take from 15 to 342 seconds per 1,000 words, OpenAI GPT-5.6 Luna the least and z.ai GLM-5 the most. Under the Basic, Standard and Advanced requirements alike, the fastest solution that meets them is on OpenAI GPT-5.5, at 120 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 |
|---|---|---|---|---|---|
| Anthropic Claude Fable 5 | one-step | 100% | ✓ | $39.95 | 259 |
| OpenAI GPT-5.5 | two-step | 100% | ✓ | $11.71 | 120 |
For tone of voice in Georgian we check the formal register, with თქვენ as the form of address. Translations from every shortlisted one-step solution, z.ai GLM-5 included, have 100% of the sentences in the formal register. Georgian has no grammatical gender for persons, so translations are neutral by default, and we check gendered terms. Translations from both of the two best solutions have 100% of the sentences about people gender-neutral without the gender-neutrality agent.
Both pairs of automatic metrics agree in Georgian, each above its threshold, and the three LQA judges agree on a median of 92.7% of the sentences of the shortlisted solutions. So the automatic rankings and the Intento LQA results on this page rest on measurement we can rely on in Georgian.
Leave feedback