Best LLMs for Language Identification
Identifies the primary language of supplied text and returns its normalized language code in the configured response schema. Illustrative uses include routing multilingual support tickets, contracts, software issues, international customer feedback, news submissions, research doc
Run this task on Fronset — request an invitation
Models
Frontier on this task: Claude Sonnet 5.5 at 10.00 / 10. Quality bar at 90%: 9.00.
point-estimate floor (CI low) · upper CI (less certain) · Bars sorted by blended cost; best-value model first.
| Model | Quality score | CI low | Cost / 1k runs | vs best value |
|---|---|---|---|---|
| Xiaomi MiMo V2.6 Flash | 9.86 / 10 | 9.71 | $0.02 | best value |
| Claude Haiku 5.5 | 9.84 / 10 | 9.68 | $0.05 | 2.2x more expensive |
| DeepSeek V4.1 Flash | 9.97 / 10 | 9.91 | $0.06 | 2.7x more expensive |
| GPT-6 Luna | 9.85 / 10 | 9.75 | $0.07 | 2.9x more expensive |
| Gemini 3.5 Flash Lite | 9.84 / 10 | 9.74 | $0.07 | 3.2x more expensive |
| NVIDIA Nemotron-3 Super 120B | 9.85 / 10 | 9.75 | $0.09 | 3.7x more expensive |
| NVIDIA Nemotron-3 Nano 30B-A3B | 9.82 / 10 | 9.74 | $0.09 | 3.9x more expensive |
| NVIDIA Nemotron 3.5 Lightning | 9.58 / 10 | 9.31 | $0.16 | 7x more expensive |
| Qwen 3.8 Flash | 9.82 / 10 | 9.68 | $0.17 | 7.5x more expensive |
| MiniMax M3 | 9.86 / 10 | 9.78 | $0.17 | 7.5x more expensive |
| Xiaomi MiMo V2.6 Pro | 9.88 / 10 | 9.78 | $0.18 | 7.8x more expensive |
| Thinking Machines Inkling Small | 9.77 / 10 | 9.61 | $0.23 | 9.9x more expensive |
| Tencent Hy3 | 9.85 / 10 | 9.78 | $0.31 | 13x more expensive |
| Gemini 3.8 Flash | 9.84 / 10 | 9.64 | $0.33 | 14x more expensive |
| GLM-5.3 Flash | 9.68 / 10 | 9.42 | $0.44 | 19x more expensive |
| NVIDIA Nemotron-3 Ultra 550B | 9.85 / 10 | 9.74 | $0.61 | 26x more expensive |
| GPT-6 Sol | 9.66 / 10 | 9.28 | $0.68 | 29x more expensive |
| Claude Sonnet 5.5 | 10.00 / 10 | 9.89 | $0.92 | 40x more expensive |
| Thinking Machines Inkling | 9.68 / 10 | 9.37 | $1.71 | 74x more expensive |
| Qwen 3.7 Plus | 9.84 / 10 | 9.70 | $1.80 | 78x more expensive |
| Claude Opus 5.5 | 9.71 / 10 | 9.35 | $1.86 | 81x more expensive |
| Tencent Hy4 Preview | 9.81 / 10 | 9.53 | $2.14 | 93x more expensive |
| Qwen 3.8 Max | 9.87 / 10 | 9.73 | $2.22 | 96x more expensive |
| Moonshot Kimi K3 | 9.95 / 10 | 9.90 | $2.79 | 121x more expensive |
| Meta Muse Spark 1.3 | 9.78 / 10 | 9.59 | $3.08 | 133x more expensive |
| Grok 4.7 | 9.84 / 10 | 9.51 | $3.12 | 135x more expensive |
| GLM-5.3 | 9.79 / 10 | 9.47 | $3.44 | 149x more expensive |
Cost breakdown
| Model | Quality | Confidence | Cost / 1k runs | Overpay | Mode |
|---|---|---|---|---|---|
| Xiaomi MiMo V2.6 Flash ★ Xiaomi MiMo | 9.86 / 10 CI [9.71, 10.00] | RANKED | $0.02 | best value | batch |
| Claude Haiku 5.5 Anthropic | 9.84 / 10 CI [9.68, 10.00] | RANKED | $0.05 | 2.2x | batch |
| DeepSeek V4.1 Flash DeepSeek | 9.97 / 10 CI [9.91, 10.00] | RANKED | $0.06 | 2.7x | batch |
| GPT-6 Luna OpenAI | 9.85 / 10 CI [9.75, 9.95] | RANKED | $0.07 | 2.9x | batch |
| Gemini 3.5 Flash Lite Gemini | 9.84 / 10 CI [9.74, 9.95] | RANKED | $0.07 | 3.2x | batch |
| NVIDIA Nemotron-3 Super 120B OpenRouter | 9.85 / 10 CI [9.75, 9.95] | RANKED | $0.09 | 3.7x | batch |
| NVIDIA Nemotron-3 Nano 30B-A3B OpenRouter | 9.82 / 10 CI [9.74, 9.89] | RANKED | $0.09 | 3.9x | batch |
| NVIDIA Nemotron 3.5 Lightning OpenRouter | 9.58 / 10 CI [9.31, 9.85] | HIGH | $0.16 | 7x | batch |
| Qwen 3.8 Flash Alibaba Cloud (DashScope) | 9.82 / 10 CI [9.68, 9.97] | RANKED | $0.17 | 7.5x | batch |
| MiniMax M3 OpenRouter | 9.86 / 10 CI [9.78, 9.94] | RANKED | $0.17 | 7.5x | batch |
| Xiaomi MiMo V2.6 Pro Xiaomi MiMo | 9.88 / 10 CI [9.78, 9.98] | RANKED | $0.18 | 7.8x | batch |
| Thinking Machines Inkling Small OpenRouter | 9.77 / 10 CI [9.61, 9.93] | RANKED | $0.23 | 9.9x | batch |
| Tencent Hy3 OpenRouter | 9.85 / 10 CI [9.78, 9.93] | RANKED | $0.31 | 13x | batch |
| Gemini 3.8 Flash Gemini | 9.84 / 10 CI [9.64, 10.00] | RANKED | $0.33 | 14x | batch |
| GLM-5.3 Flash Z.AI | 9.68 / 10 CI [9.42, 9.95] | HIGH | $0.44 | 19x | batch |
| NVIDIA Nemotron-3 Ultra 550B OpenRouter | 9.85 / 10 CI [9.74, 9.96] | RANKED | $0.61 | 26x | batch |
| GPT-6 Sol OpenAI | 9.66 / 10 CI [9.28, 10.00] | MEDIUM | $0.68 | 29x | batch |
| Claude Sonnet 5.5 best Anthropic | 10.00 / 10 CI [9.89, 10.00] | RANKED | $0.92 | 40x | batch |
| Thinking Machines Inkling OpenRouter | 9.68 / 10 CI [9.37, 9.99] | MEDIUM | $1.71 | 74x | batch |
| Qwen 3.7 Plus Alibaba Cloud (DashScope) | 9.84 / 10 CI [9.70, 9.98] | RANKED | $1.80 | 78x | batch |
| Claude Opus 5.5 Anthropic | 9.71 / 10 CI [9.35, 10.00] | MEDIUM | $1.86 | 81x | batch |
| Tencent Hy4 Preview OpenRouter | 9.81 / 10 CI [9.53, 10.00] | HIGH | $2.14 | 93x | batch |
| Qwen 3.8 Max Alibaba Cloud (DashScope) | 9.87 / 10 CI [9.73, 10.00] | RANKED | $2.22 | 96x | batch |
| Moonshot Kimi K3 Moonshot AI | 9.95 / 10 CI [9.90, 10.00] | RANKED | $2.79 | 121x | batch |
| Meta Muse Spark 1.3 OpenRouter | 9.78 / 10 CI [9.59, 9.97] | RANKED | $3.08 | 133x | batch |
| Grok 4.7 xAI | 9.84 / 10 CI [9.51, 10.00] | MEDIUM | $3.12 | 135x | batch |
| GLM-5.3 Z.AI | 9.79 / 10 CI [9.47, 10.00] | MEDIUM | $3.44 | 149x | batch |
Overpay shows how much more you pay than the best-value model that clears the quality bar (marked ★) — the best-value good-enough option. "16x" means you overpay 16× — 16× that reference for no quality benefit above the bar. Typical call shape for this task: 439 input tokens → 117 output tokens, EMA-tracked from production traffic. Cost is the observed, all-in $ per 1,000 task runs: each model's own measured usage on this task — output verbosity, thinking/reasoning tokens, cache reads and writes, and the spend on its billed failures — priced at current list rates and adjusted by the billing overhead we actually reconcile against provider invoices. Models that answer tersely cost what they actually cost; models that think at length pay for it. Not comparable to providers' advertised $/1M list rates — this is what running the task costs, not a per-token price.
Evaluation rubric
Judge primary-language correctness, handling of mixed or noisy text, correct normalization to the requested code set, and calibrated uncertainty. Formatting is deterministic.
Output schema
Every answer on this task is checked against this JSON Schema, whichever model wrote it. An answer that doesn't fit counts as a model failure, and the call is retried on another model.
{
"description": "Schema for the result of language detection.",
"properties": {
"language_code": {
"description": "Detected language code (e.g., 'en', 'es', 'fr'). Should be 'unknown' if not confidently detected.",
"title": "Language Code",
"type": "string"
}
},
"required": [
"language_code"
],
"title": "LanguageDetectionResult",
"type": "object"
}Prompt templates
The published template for this task.
default ·
LLMB_LANGUAGE_IDENTIFICATION_SYSTEM +
LLMB_LANGUAGE_IDENTIFICATION_USER
(98740 calls in window)
System prompt
Identify the language that carries most of the semantic content. Ignore URLs, handles, code fragments, boilerplate, and isolated borrowed words. When text is too short, mixed, or ambiguous, use the schema’s unknown or best-supported behavior rather than fabricating confidence. Return the requested code standard only. Treat empty optional values as absent and return only the requested result. Your response must conform exactly to this output schema: {schema_json_string}.
User prompt
Inputs — input_text: {input_text}. Use only these inputs to complete the task defined by the system prompt.