Fronset Benchmark

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

Cost mode:

Category: Relevance, Classification & Matching · Typical I/O: 652→37 tokens

Runs on model group: All models. Inputs may reach any model in the group, not only the ranked models below.

Run this task on Fronset — request an invitation

Models

Frontier on this task: Xiaomi MiMo V2.6 Flash at 10.14 / 10. Quality bar at 90%: 9.12.

point-estimate floor (CI low) · upper CI (less certain) · Bars sorted by blended cost; best-value model first.

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
Xiaomi MiMo V2.6 Flash ★ best Xiaomi MiMo10.14 / 10 CI [9.82, 10.00]MEDIUM$0.03best valuebatch
GPT-6 Luna OpenAI10.00 / 10 CI [9.98, 10.00]RANKED$0.041.4xbatch
Gemini 3.5 Flash Lite Gemini9.91 / 10 CI [9.79, 10.00]RANKED$0.072.6xbatch
NVIDIA Nemotron-3 Nano 30B-A3B OpenRouter9.89 / 10 CI [9.81, 9.97]RANKED$0.082.8xbatch
NVIDIA Nemotron-3 Super 120B OpenRouter9.92 / 10 CI [9.81, 10.00]RANKED$0.103.5xbatch
Xiaomi MiMo V2.6 Pro Xiaomi MiMo10.05 / 10 CI [9.91, 10.00]RANKED$0.186.5xbatch
MiniMax M3 OpenRouter9.98 / 10 CI [9.89, 10.00]RANKED$0.196.7xbatch
NVIDIA Nemotron 3.5 Lightning OpenRouter9.59 / 10 CI [9.24, 9.94]MEDIUM$0.206.9xbatch
Thinking Machines Inkling Small OpenRouter9.84 / 10 CI [9.66, 10.00]RANKED$0.248.4xbatch
Tencent Hy3 OpenRouter9.95 / 10 CI [9.87, 10.00]RANKED$0.3211xbatch
Qwen 3.8 Flash Alibaba Cloud (DashScope)9.88 / 10 CI [9.72, 10.00]RANKED$0.3412xbatch
Gemini 3.8 Flash Gemini9.92 / 10 CI [9.77, 10.00]RANKED$0.3713xbatch
GLM-5.3 Flash Z.AI9.77 / 10 CI [9.53, 10.00]HIGH$0.4416xbatch
GPT-6 Sol OpenAI9.68 / 10 CI [9.22, 10.00]MEDIUM$0.6322xbatch
NVIDIA Nemotron-3 Ultra 550B OpenRouter9.95 / 10 CI [9.82, 10.00]RANKED$0.7326xbatch
Qwen 3.7 Plus Alibaba Cloud (DashScope)9.90 / 10 CI [9.76, 10.00]RANKED$1.0838xbatch
Claude Opus 5.5 Anthropic9.80 / 10 CI [9.39, 10.00]MEDIUM$1.7863xbatch
Thinking Machines Inkling OpenRouter9.76 / 10 CI [9.41, 10.00]MEDIUM$1.9468xbatch
Tencent Hy4 Preview OpenRouter9.91 / 10 CI [9.70, 10.00]HIGH$2.1676xbatch
Qwen 3.8 Max Alibaba Cloud (DashScope)10.06 / 10 CI [9.96, 10.00]RANKED$2.3483xbatch
Moonshot Kimi K3 Moonshot AI10.04 / 10 CI [10.01, 10.00]RANKED$2.7095xbatch
GLM-5.3 Z.AI9.90 / 10 CI [9.68, 10.00]HIGH$3.50123xbatch
Meta Muse Spark 1.3 OpenRouter9.88 / 10 CI [9.61, 10.00]HIGH$3.67129xbatch

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: 652 input tokens → 37 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 (79871 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.