The State of Translation Automation 2026 · Languages
What it takes to automate translation from English into Tagalog 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 Tagalog. |
| Tone of voice | The formal register (po, opo, kayo) 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 Tagalog, 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.
Four three-step solutions (translation + LQA + repair) meet every Standard requirement in Tagalog. None of them leads on errors, cost and time at once. Optimize for selects OpenAI GPT-5.5 for quality and OpenAI GPT-5.6 Terra for cost and latency. The one on Anthropic Claude Fable 5 has 11.9 total errors per 1,000 words and takes 216 seconds per 1,000 words, but costs the most, $636 per million characters. The one on OpenAI GPT-5.6 Terra has fewer errors, a lower cost and a shorter time than the one on Anthropic Claude Sonnet 4.6.
OpenAI GPT-5.5
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.5
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.6 Terra
three-step
OpenAI GPT-5.5
three-step
OpenAI GPT-5.5
three-step
Anthropic Claude Fable 5
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 | ☆×5 | three-step | OpenAI GPT-5.6 Terra | 12.2 | $541 | 165 |
| Standard | Tier 2: two or more solutions, none simple | ☆×4 | three-step | OpenAI GPT-5.6 Terra | 12.2 | $541 | 165 |
| Advanced | Tier 2: two or more solutions, none simple | ☆☆ | three-step | OpenAI GPT-5.5 | 7.6 | $566 | 240 |
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.
The Standard requirements remove one of the five three-step solutions (translation + LQA + repair) that meet every Basic requirement in Tagalog, the one on Google Gemini 3.1 Pro Preview. Translations from it have 97.9% of the sentences in the formal register, against a bar of 100%. Under the Basic and Standard requirements, Optimize for selects OpenAI GPT-5.5 for quality and OpenAI GPT-5.6 Terra for cost and latency. Under the Advanced ones, it selects OpenAI GPT-5.5 for quality and cost and Anthropic Claude Fable 5 for latency.
| Model | US dollars per million characters | Weighted errors per 1,000 words | Group |
|---|---|---|---|
| Google Gemma 4 | 0.15 | 35.9 | Misses a Standard requirement |
| Mistral Large 3 | 0.86 | 72.8 | Misses a Standard requirement |
| z.ai GLM-5 | 2.13 | 53.4 | Misses a Standard requirement |
| Google Gemini 3.5 Flash | 3.79 | 45.2 | Misses a Standard requirement |
| OpenAI GPT-5.6 Terra | 4.52 | 21.4 | Misses a Standard requirement |
| Google Gemini 3.1 Pro Preview | 5.12 | 40.9 | Misses a Standard requirement |
| Anthropic Claude Sonnet 4.6 | 7.37 | 44.5 | Misses a Standard requirement |
| OpenAI GPT-5.5 | 11.23 | 23.7 | Misses a Standard requirement |
| Google Translation LLM | 22.41 | 78.4 | Misses a Standard requirement |
| Translated Lara 2 | 24.99 | 45.8 | Misses a Standard requirement |
| DeepL Next Gen | 27.5 | 55.4 | Misses a Standard requirement |
| Anthropic Claude Fable 5 | 33.06 | 29 | Misses a Standard requirement |
Among the one-step solutions in Tagalog, the one on OpenAI GPT-5.6 Terra has the fewest weighted errors, 21.4 per 1,000 words, at $4.52 per million characters. Every one-step solution that costs more has more errors. The solutions on the three LLM-based MT engines cost $22.41 to $27.50 per million characters and have 45.8 to 78.4 weighted errors per 1,000 words. The cheapest, on the open-weight Google Gemma 4, costs $0.15 per million characters and has 35.9 weighted errors per 1,000 words.
| Requirement | Median errors per 1,000 words |
|---|---|
| General translation quality | 44.9 |
| Tone of voice | 3.1 |
| Tag handling | 0 |
| Gender neutrality | 0 |
The requirements with countable errors. Every solution with its result on each requirement: the explorer.
None of the one-step or two-step solutions (translation + post-edit) meet general translation quality in Tagalog. The one-step solutions on Google Gemini 3.1 Pro Preview, Google Gemini 3.5 Flash and z.ai GLM-5 meet every other Standard requirement, with 40.9, 45.2 and 53.4 weighted errors per 1,000 words. Three-step solutions (translation + LQA + repair) bring five models under the bar, at 7.6 to 13.5 weighted errors per 1,000 words. On four of them, 100% of the sentences are in the formal register, against 93.6% to 97.9% in one step.
| Model | Weighted errors per 1,000 words of the shortlisted one-step solutions in Tagalog | Class |
|---|---|---|
| Google Translation LLM | 78.4 | LLM-based MT engines |
| Mistral Large 3 | 72.8 | LLMs |
| DeepL Next Gen | 55.4 | LLM-based MT engines |
| z.ai GLM-5 | 53.4 | LLMs |
| Translated Lara 2 | 45.8 | LLM-based MT engines |
| Google Gemini 3.5 Flash | 45.2 | LLMs |
| Anthropic Claude Sonnet 4.6 | 44.5 | LLMs |
| Google Gemini 3.1 Pro Preview | 40.9 | LLMs |
| Google Gemma 4 | 35.9 | Open-weight LLMs |
| Anthropic Claude Fable 5 | 29 | LLMs |
| OpenAI GPT-5.5 | 23.7 | LLMs |
| OpenAI GPT-5.6 Terra | 21.4 | LLMs |
In one step, translations from the eight shortlisted LLMs other than the open-weight one have 21.4 to 72.8 weighted errors per 1,000 words in Tagalog, and translations from the open-weight Google Gemma 4 have 35.9. Translations from Google Translation LLM have 9 critical errors, the only critical errors among the shortlisted one-step solutions. The shortlisted LLM that does not meet our language-support criteria in Tagalog is Mistral Large 3.
| Model | COMET-Kiwi | Rank on COMET-Kiwi | MetricX | Rank on MetricX | Shortlisted | Shortlist place |
|---|---|---|---|---|---|---|
| Anthropic Claude Opus 4.8 | 0.783 | 9 | 2.11 | 12 | No | Not shortlisted |
| OpenAI GPT-4.1 | 0.781 | 10 | 2.18 | 14 | No | Not shortlisted |
| xAI Grok 4.20 (reasoning) | 0.781 | 11 | 2.24 | 16 | No | Not shortlisted |
| Anthropic Claude Sonnet 5 | 0.781 | 12 | 2.27 | 19 | No | Not shortlisted |
| MoonshotAI Kimi K2.5 | 0.779 | 13 | 2.41 | 20 | No | Not shortlisted |
| DeepSeek V3.2 | 0.777 | 15 | 1.99 | 9 | No | Not shortlisted |
| Meta Llama 4 Maverick | 0.775 | 18 | 2.09 | 11 | No | Not shortlisted |
| Cohere Command A Translate | 0.774 | 19 | 2.19 | 15 | No | Not shortlisted |
| OpenAI gpt-oss-120b | 0.767 | 21 | 2.57 | 23 | No | Not shortlisted |
| Mistral Medium 3.1 | 0.766 | 22 | 2.26 | 18 | No | Not shortlisted |
| Alibaba Qwen3-235B-A22B | 0.757 | 23 | 2.56 | 22 | No | Not shortlisted |
| MiniMax M2.5 | 0.756 | 24 | 2.99 | 28 | No | Not shortlisted |
| Cohere Tiny Aya | 0.755 | 25 | 2.79 | 24 | No | Not shortlisted |
| Google Cloud Advanced Translation API | 0.755 | 26 | 2.95 | 26 | No | Not shortlisted |
| Swiss AI Apertus-8B-2509 | 0.750 | 27 | 2.87 | 25 | No | Not shortlisted |
| AI21 Jamba Large 1.7 | 0.741 | 28 | 3.06 | 29 | No | Not shortlisted |
| Anthropic Claude Haiku 4.5 | 0.740 | 29 | 3.78 | 34 | No | Not shortlisted |
| Yandex Cloud Translate v2 | 0.740 | 30 | 2.98 | 27 | No | Not shortlisted |
| ModernMT Enterprise Edition | 0.732 | 31 | 3.21 | 31 | No | Not shortlisted |
| SYSTRAN PNMT API | 0.723 | 32 | 3.26 | 32 | No | Not shortlisted |
| Amazon Translate | 0.714 | 33 | 3.18 | 30 | No | Not shortlisted |
| Alibaba Machine Translation API | 0.694 | 34 | 3.51 | 33 | No | Not shortlisted |
| AppTek NMT | 0.616 | 35 | 6.04 | 35 | No | Not shortlisted |
| Anthropic Claude Sonnet 4.6 | 0.794 | 1 | 2.01 | 10 | Yes | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Terra | 0.793 | 2 | 1.89 | 4 | Yes | Top 8 by both metrics |
| OpenAI GPT-5.5 | 0.793 | 3 | 1.86 | 3 | Yes | Top 8 by both metrics |
| Anthropic Claude Fable 5 | 0.789 | 4 | 2.18 | 13 | Yes | Top 8 by COMET-Kiwi |
| Google Gemma 4 | 0.789 | 5 | 1.97 | 6 | Yes | Top 8 by both metrics |
| Google Gemini 3.5 Flash | 0.787 | 6 | 1.97 | 7 | Yes | Top 8 by both metrics |
| DeepL Next Gen | 0.787 | 7 | 1.89 | 5 | Yes | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | 0.785 | 8 | 1.83 | 1 | Yes | Top 8 by both metrics |
| Translated Lara 2 | 0.777 | 14 | 2.25 | 17 | Yes | Real-time place |
| Google Translation LLM | 0.777 | 16 | 2.51 | 21 | Yes | Real-time place |
| Mistral Large 3 | 0.777 | 17 | 1.83 | 2 | Yes | Top 8 by MetricX |
| z.ai GLM-5 | 0.767 | 20 | 1.98 | 8 | Yes | Top 8 by MetricX |
| Model | Solution complexity | COMET-Kiwi rank | COMET-Kiwi | MetricX rank | MetricX | Shortlist |
|---|---|---|---|---|---|---|
| Anthropic Claude Sonnet 4.6 | Adjusted LLM | 1 | 0.794 | 10 | 2.01 | Top 8 by COMET-Kiwi |
| OpenAI GPT-5.6 Terra | Adjusted LLM | 2 | 0.793 | 4 | 1.89 | Top 8 by both metrics |
| OpenAI GPT-5.5 | Adjusted LLM | 3 | 0.793 | 3 | 1.86 | Top 8 by both metrics |
| Anthropic Claude Fable 5 | Adjusted LLM | 4 | 0.789 | 13 | 2.18 | Top 8 by COMET-Kiwi |
| Google Gemma 4 | Adjusted LLM | 5 | 0.789 | 6 | 1.97 | Top 8 by both metrics |
| Google Gemini 3.5 Flash | Adjusted LLM | 6 | 0.787 | 7 | 1.97 | Top 8 by both metrics |
| DeepL Next Gen | Off-the-shelf NMT | 7 | 0.787 | 5 | 1.89 | Top 8 by both metrics |
| Google Gemini 3.1 Pro Preview | Adjusted LLM | 8 | 0.785 | 1 | 1.83 | Top 8 by both metrics |
| Anthropic Claude Opus 4.8 | Adjusted LLM | 9 | 0.783 | 12 | 2.11 | Not shortlisted |
| OpenAI GPT-4.1 | Adjusted LLM | 10 | 0.781 | 14 | 2.18 | Not shortlisted |
| xAI Grok 4.20 (reasoning) | Adjusted LLM | 11 | 0.781 | 16 | 2.24 | Not shortlisted |
| Anthropic Claude Sonnet 5 | Adjusted LLM | 12 | 0.781 | 19 | 2.27 | Not shortlisted |
| MoonshotAI Kimi K2.5 | Adjusted LLM | 13 | 0.779 | 20 | 2.41 | Not shortlisted |
| Translated Lara 2 | Adjusted NMT | 14 | 0.777 | 17 | 2.25 | Real-time place |
| DeepSeek V3.2 | Adjusted LLM | 15 | 0.777 | 9 | 1.99 | Not shortlisted |
| Google Translation LLM | Adjusted NMT | 16 | 0.777 | 21 | 2.51 | Real-time place |
| Mistral Large 3 | Adjusted LLM | 17 | 0.777 | 2 | 1.83 | Top 8 by MetricX |
| Meta Llama 4 Maverick | Adjusted LLM | 18 | 0.775 | 11 | 2.09 | Not shortlisted |
| Cohere Command A Translate | Adjusted LLM | 19 | 0.774 | 15 | 2.19 | Not shortlisted |
| z.ai GLM-5 | Adjusted LLM | 20 | 0.767 | 8 | 1.98 | Top 8 by MetricX |
| OpenAI gpt-oss-120b | Off-the-shelf LLM | 21 | 0.767 | 23 | 2.57 | Not shortlisted |
| Mistral Medium 3.1 | Adjusted LLM | 22 | 0.766 | 18 | 2.26 | Not shortlisted |
| Alibaba Qwen3-235B-A22B | Adjusted LLM | 23 | 0.757 | 22 | 2.56 | Not shortlisted |
| MiniMax M2.5 | Adjusted LLM | 24 | 0.756 | 28 | 2.99 | Not shortlisted |
| Cohere Tiny Aya | Off-the-shelf LLM | 25 | 0.755 | 24 | 2.79 | Not shortlisted |
| Google Cloud Advanced Translation API | Off-the-shelf NMT | 26 | 0.755 | 26 | 2.95 | Not shortlisted |
| Swiss AI Apertus-8B-2509 | Adjusted LLM | 27 | 0.750 | 25 | 2.87 | Not shortlisted |
| AI21 Jamba Large 1.7 | Adjusted LLM | 28 | 0.741 | 29 | 3.06 | Not shortlisted |
| Anthropic Claude Haiku 4.5 | Adjusted LLM | 29 | 0.740 | 34 | 3.78 | Not shortlisted |
| Yandex Cloud Translate v2 | Off-the-shelf NMT | 30 | 0.740 | 27 | 2.98 | Not shortlisted |
| ModernMT Enterprise Edition | Off-the-shelf NMT | 31 | 0.732 | 31 | 3.21 | Not shortlisted |
| SYSTRAN PNMT API | Off-the-shelf NMT | 32 | 0.723 | 32 | 3.26 | Not shortlisted |
| Amazon Translate | Off-the-shelf NMT | 33 | 0.714 | 30 | 3.18 | Not shortlisted |
| Alibaba Machine Translation API | Off-the-shelf NMT | 34 | 0.694 | 33 | 3.51 | Not shortlisted |
| AppTek NMT | Off-the-shelf NMT | 35 | 0.616 | 35 | 6.04 | Not shortlisted |
| OpenAI GPT-5.6 Luna | Adjusted LLM | — | 0.793 | — | 1.86 | Same model family as OpenAI GPT-5.6 Terra |
| OpenAI GPT-5.6 Sol | Adjusted LLM | — | 0.791 | — | 1.70 | Same model family as OpenAI GPT-5.6 Terra |
| xAI Grok 4.20 (non-reasoning) | Adjusted LLM | — | 0.775 | — | 2.32 | 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 35 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.
By Intento LQA, the one-step solutions with the fewest weighted errors in Tagalog are on OpenAI GPT-5.6 Terra and OpenAI GPT-5.5, both in the top 4 of each metric. The one on Anthropic Claude Sonnet 4.6, first on COMET-Kiwi, has 44.5 weighted errors per 1,000 words, and the one on Google Gemini 3.1 Pro Preview, first on MetricX, 40.9. Anthropic Claude Fable 5 is on the shortlist through COMET-Kiwi alone, 13th on MetricX. Nearest the cut in the tail are Anthropic Claude Opus 4.8, 9th on COMET-Kiwi, and DeepSeek V3.2, 9th 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 | $541 |
| Standard | OpenAI GPT-5.6 Terra, three-step | $541 |
| Advanced | OpenAI GPT-5.5, three-step | $566 |
| Model | US dollars per million characters | Class |
|---|---|---|
| Google Gemma 4 | $0.15 | Open-weight LLMs |
| Mistral Large 3 | $0.86 | LLMs |
| z.ai GLM-5 | $2.13 | LLMs |
| Google Gemini 3.5 Flash | $3.79 | LLMs |
| OpenAI GPT-5.6 Terra | $4.52 | LLMs |
| Google Gemini 3.1 Pro Preview | $5.12 | LLMs |
| Anthropic Claude Sonnet 4.6 | $7.37 | LLMs |
| OpenAI GPT-5.5 | $11.23 | LLMs |
| Google Translation LLM | $22.41 | 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 | $33.06 | LLMs |
| Model | Tokens per character | Class |
|---|---|---|
| Anthropic Claude Opus 4.8 | 0.485 | LLMs |
| Anthropic Claude Sonnet 5 | 0.485 | LLMs |
| Anthropic Claude Fable 5 | 0.482 | LLMs |
| Mistral Large 3 | 0.387 | LLMs |
| Mistral Medium 3.1 | 0.387 | LLMs |
| Anthropic Claude Sonnet 4.6 | 0.362 | LLMs |
| AI21 Jamba Large 1.7 | 0.348 | LLMs |
| Anthropic Claude Haiku 4.5 | 0.348 | LLMs |
| Alibaba Qwen3-235B-A22B | 0.335 | Open-weight LLMs |
| Cohere Command A Translate | 0.332 | LLMs |
| z.ai GLM-5 | 0.328 | LLMs |
| DeepSeek V3.2 | 0.326 | LLMs |
| MoonshotAI Kimi K2.5 | 0.323 | LLMs |
| xAI Grok 4.20 (reasoning) | 0.311 | LLMs |
| Google Gemini 3.5 Flash | 0.306 | LLMs |
| Google Gemini 3.1 Pro Preview | 0.304 | LLMs |
| Google Gemma 4 | 0.304 | Open-weight LLMs |
| xAI Grok 4.20 (non-reasoning) | 0.301 | LLMs |
| MiniMax M2.5 | 0.28 | LLMs |
| OpenAI GPT-5.6 Luna | 0.28 | LLMs |
| OpenAI GPT-5.6 Sol | 0.28 | LLMs |
| OpenAI GPT-5.6 Terra | 0.28 | LLMs |
| OpenAI GPT-4.1 | 0.275 | LLMs |
| OpenAI GPT-5.5 | 0.275 | LLMs |
| Cohere Tiny Aya | 0.264 | LLMs |
The tokenizer of Anthropic Claude Fable 5 produces 0.482 tokens per character of the Tagalog translation, against 0.275 for OpenAI GPT-5.5. The one-step solution on Anthropic Claude Fable 5 costs $33.06 per million characters, the most on the shortlist, against $11.23 on OpenAI GPT-5.5. Tokens alone do not set the cost. The tokenizer of Anthropic Claude Sonnet 4.6 produces 0.362 tokens per character, and the one-step solution on it costs $7.37 per million characters.
| Quality level | Fastest solution that meets every requirement of the level | Seconds per 1,000 words |
|---|---|---|
| Basic | OpenAI GPT-5.6 Terra, three-step | 165 |
| Standard | OpenAI GPT-5.6 Terra, three-step | 165 |
| Advanced | Anthropic Claude Fable 5, three-step | 216 |
| Model | Seconds per 1,000 words | Class |
|---|---|---|
| Google Cloud Advanced Translation API | 0.1 | MT engines |
| ModernMT Enterprise Edition | 0.3 | MT engines |
| DeepL Next Gen | 0.6 | LLM-based MT engines |
| Google Translation LLM | 0.6 | LLM-based MT engines |
| SYSTRAN PNMT API | 0.7 | MT engines |
| Amazon Translate | 1.3 | MT engines |
| Translated Lara 2 | 1.5 | LLM-based MT engines |
| Yandex Cloud Translate v2 | 2.3 | MT engines |
| OpenAI gpt-oss-120b | 3.2 | Open-weight LLMs |
| Google Gemini 3.5 Flash | 16 | LLMs |
| OpenAI GPT-5.6 Luna | 17 | LLMs |
| xAI Grok 4.20 (non-reasoning) | 18 | LLMs |
| Cohere Tiny Aya | 21 | LLMs |
| Alibaba Machine Translation API | 26 | MT engines |
| OpenAI GPT-5.6 Terra | 34 | LLMs |
| Anthropic Claude Haiku 4.5 | 36 | LLMs |
| OpenAI GPT-4.1 | 39 | LLMs |
| Meta Llama 4 Maverick | 45 | Open-weight LLMs |
| MiniMax M2.5 | 48 | LLMs |
| Anthropic Claude Sonnet 5 | 51 | LLMs |
| DeepSeek V3.2 | 51 | LLMs |
| xAI Grok 4.20 (reasoning) | 53 | LLMs |
| Cohere Command A Translate | 55 | LLMs |
| AI21 Jamba Large 1.7 | 56 | LLMs |
| Anthropic Claude Fable 5 | 58 | LLMs |
| Mistral Medium 3.1 | 59 | LLMs |
| Anthropic Claude Opus 4.8 | 63 | LLMs |
| OpenAI GPT-5.5 | 71 | LLMs |
| Anthropic Claude Sonnet 4.6 | 74 | LLMs |
| Mistral Large 3 | 76 | LLMs |
| Google Gemini 3.1 Pro Preview | 78 | LLMs |
| Swiss AI Apertus-8B-2509 | 83 | Open-weight LLMs |
| OpenAI GPT-5.6 Sol | 84 | LLMs |
| Alibaba Qwen3-235B-A22B | 137 | Open-weight LLMs |
| AppTek NMT | 139 | MT engines |
| Google Gemma 4 | 185 | Open-weight LLMs |
| MoonshotAI Kimi K2.5 | 245 | LLMs |
| z.ai GLM-5 | 266 | 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 Tagalog.
The eight real-time models in Tagalog are four MT engines, three LLM-based MT engines and the open-weight OpenAI gpt-oss-120b. The three on the shortlist, DeepL Next Gen, Translated Lara 2 and Google Translation LLM, are the LLM-based ones. None of them meet general translation quality, tone of voice or formatting. Of the three, only translations from Translated Lara 2 meet tag handling. On the shortlist, the one-step solutions on the LLMs other than the open-weight Google Gemma 4 take 16 to 266 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 |
|---|---|---|---|---|---|
| OpenAI GPT-5.5 | three-step | 100% | ✓ | $566 | 240 |
| Anthropic Claude Fable 5 | three-step | 100% | ✓ | $636 | 216 |
For tone of voice in Tagalog we check the formal register, with po, opo and kayo as the form of address. In one step, translations from the shortlisted LLMs have 93.6% to 100% of the sentences in that register, and from the LLM-based MT engines 8.51% to 68.1%. For gender neutrality we check that a translation does not add a gender the source does not specify, with no default gender. Tagalog has no grammatical gender for persons, so a translation is neutral by default and only has to avoid gendered terms.
For the 12 shortlisted solutions in Tagalog, the median share of sentences on which the three LQA judges agree is 94.9%. With agreement that high, the Intento LQA error counts on this page are reliable.
Leave feedback