Cost mode:

Category: Infrastructure & Utility · Typical I/O: 1008→1298 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: Moonshot Kimi K3 at 9.23 / 10. Quality bar at 90%: 8.31.

point-estimate floor (CI low) · upper CI (less certain) · Bars sorted by blended cost; best-value model first. Greyed rows are MEDIUM+ models whose point estimate does not clear the bar.

ModelQuality scoreCI lowCost / 1k runsvs best value
GPT-6 Luna8.72 / 108.45$0.33best value
Xiaomi MiMo V2.6 Pro8.93 / 108.70$0.561.7x more expensive
Gemini 3.5 Flash Lite8.59 / 108.28$0.591.8x more expensive
MiniMax M38.79 / 108.57$1.023.1x more expensive
DeepSeek V4.1 Flash8.83 / 108.54$2.206.6x more expensive
Tencent Hy38.36 / 108.06$2.427.3x more expensive
Gemini 3.8 Flash9.01 / 108.77$3.6611x more expensive
Thinking Machines Inkling Small8.48 / 108.13$4.2913x more expensive
GPT-6 Sol9.01 / 108.79$4.6914x more expensive
Qwen 3.7 Plus8.92 / 108.78$5.3716x more expensive
Claude Sonnet 5.59.02 / 108.72$5.7717x more expensive
GLM-5.3 Flash8.92 / 108.67$6.0118x more expensive
Meta Muse Spark 1.38.92 / 108.73$6.5620x more expensive
Claude Opus 5.59.12 / 108.93$11.3734x more expensive
Thinking Machines Inkling8.59 / 108.18$14.9645x more expensive
Moonshot Kimi K39.23 / 109.07$30.9493x more expensive
Grok 4.78.71 / 108.31$34.83105x more expensive
Qwen 3.8 Max8.49 / 108.11$35.62107x more expensive
GLM-5.38.92 / 108.61$69.48209x more expensive
NVIDIA Nemotron-3 Nano 30B-A3B7.32 / 106.91$0.662x more expensive
NVIDIA Nemotron-3 Super 120B8.00 / 107.62$2.296.9x more expensive
Xiaomi MiMo V2.6 Flash8.31 / 107.89$0.0972% cheaper
Qwen 3.8 Flash8.15 / 107.70$3.5411x more expensive

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
GPT-6 Luna ★ OpenAI8.72 / 10 CI [8.45, 8.98]HIGH$0.33best valuebatch
Xiaomi MiMo V2.6 Pro Xiaomi MiMo8.93 / 10 CI [8.70, 9.15]HIGH$0.561.7xbatch
Gemini 3.5 Flash Lite Gemini8.59 / 10 CI [8.28, 8.90]MEDIUM$0.591.8xbatch
MiniMax M3 OpenRouter8.79 / 10 CI [8.57, 9.01]HIGH$1.023.1xbatch
DeepSeek V4.1 Flash DeepSeek8.83 / 10 CI [8.54, 9.11]HIGH$2.206.6xbatch
Tencent Hy3 OpenRouter8.36 / 10 CI [8.06, 8.67]MEDIUM$2.427.3xbatch
Gemini 3.8 Flash Gemini9.01 / 10 CI [8.77, 9.25]HIGH$3.6611xbatch
Thinking Machines Inkling Small OpenRouter8.48 / 10 CI [8.13, 8.82]MEDIUM$4.2913xbatch
GPT-6 Sol OpenAI9.01 / 10 CI [8.79, 9.23]HIGH$4.6914xbatch
Qwen 3.7 Plus Alibaba Cloud (DashScope)8.92 / 10 CI [8.78, 9.06]RANKED$5.3716xbatch
Claude Sonnet 5.5 Anthropic9.02 / 10 CI [8.72, 9.31]HIGH$5.7717xbatch
GLM-5.3 Flash Z.AI8.92 / 10 CI [8.67, 9.18]HIGH$6.0118xbatch
Meta Muse Spark 1.3 OpenRouter8.92 / 10 CI [8.73, 9.10]RANKED$6.5620xbatch
Claude Opus 5.5 Anthropic9.12 / 10 CI [8.93, 9.32]RANKED$11.3734xbatch
Thinking Machines Inkling OpenRouter8.59 / 10 CI [8.18, 9.01]MEDIUM$14.9645xbatch
Moonshot Kimi K3 best Moonshot AI9.23 / 10 CI [9.07, 9.39]RANKED$30.9493xbatch
Grok 4.7 xAI8.71 / 10 CI [8.31, 9.11]MEDIUM$34.83105xbatch
Qwen 3.8 Max Alibaba Cloud (DashScope)8.49 / 10 CI [8.11, 8.87]MEDIUM$35.62107xbatch
GLM-5.3 Z.AI8.92 / 10 CI [8.61, 9.24]MEDIUM$69.48209xbatch

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: 1008 input tokens → 1298 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 semantic fidelity, fluency in the target language, preservation of tone and qualifiers, exact handling of protected tokens, item alignment, and avoidance of added or omitted meaning.

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.

{
  "$defs": {
    "BatchTranslationItem": {
      "description": "One translated item in a batch translation result.",
      "properties": {
        "index": {
          "description": "0-based index of this item in the input list. Must match the input order.",
          "title": "Index",
          "type": "integer"
        },
        "translated_text": {
          "description": "Translated text for the item at this index.",
          "title": "Translated Text",
          "type": "string"
        }
      },
      "required": [
        "index",
        "translated_text"
      ],
      "title": "BatchTranslationItem",
      "type": "object"
    }
  },
  "description": "Batched translation: translate a list of texts in a single LLM call.\n\nThe output items list MUST be the same length as the input list, with index\nvalues 0..N-1 in order. Each translated_text corresponds to the input at the\nsame index.",
  "properties": {
    "items": {
      "description": "Translations in the same order as the input list. Length must equal the input list length.",
      "items": {
        "$ref": "#/$defs/BatchTranslationItem"
      },
      "title": "Items",
      "type": "array"
    }
  },
  "required": [
    "items"
  ],
  "title": "BatchTranslationResult",
  "type": "object"
}

Prompt templates

The published template for this task.

default · LLMB_BATCH_TEXT_TRANSLATION_SYSTEM + LLMB_BATCH_TEXT_TRANSLATION_USER (7381 calls in window)

System prompt

Translate meaning rather than word order. Preserve names, numbers, dates, URLs, handles, hashtags, code, markup, and caller-defined protected tokens exactly unless transliteration is requested. Keep item order and identifiers, reproduce ambiguity rather than resolving it with new facts, and return only the configured output language and schema. 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 — source_language: {source_language}; target_language: {target_language}; items_json: {items_json}; input_text: {input_text}. Use only these inputs to complete the task defined by the system prompt.