Cost mode:

Category: Infrastructure & Utility · Rail: absolute · Typical I/O: 905→1440 tokens

Models

Frontier on this task: Gemini 3.5 Flash at 9.38 / 10. Quality bar at 90%: 8.44.

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 clears the bar but whose CI low does not.

ModelQuality scoreCI lowCost / 1k runsvs best value
GPT-5.6 Luna8.73 / 108.55$0.34best value
Gemini 3.5 Flash Lite8.58 / 108.24$0.661.9x more expensive
MiniMax M38.81 / 108.59$1.544.5x more expensive
GLM-5.3 Flash8.89 / 108.62$2.717.9x more expensive
Gemini 3.8 Flash9.01 / 108.73$3.6611x more expensive
Qwen 3.7 Plus8.93 / 108.79$4.7714x more expensive
GPT-5.6 Terra8.98 / 108.84$4.8814x more expensive
DeepSeek V4 Flash8.53 / 108.27$5.1415x more expensive
Thinking Machines Inkling Small8.51 / 108.16$5.2115x more expensive
Meta Muse Spark 1.39.00 / 108.82$6.9020x more expensive
GPT-5.6 Sol8.93 / 108.77$7.4522x more expensive
Gemini 3.5 Flash9.38 / 109.16$8.1224x more expensive
DeepSeek V4 Pro8.54 / 108.24$9.9529x more expensive
Claude Sonnet 58.95 / 108.83$10.6231x more expensive
Thinking Machines Inkling8.59 / 108.17$12.6037x more expensive
Claude Opus 59.05 / 108.92$19.3456x more expensive
GLM-5.38.83 / 108.49$31.0991x more expensive
Tencent Hy4 Preview8.64 / 108.18$31.5992x more expensive
Grok 4.69.04 / 108.92$32.5995x more expensive
Moonshot Kimi K39.25 / 109.08$38.79113x more expensive
Qwen 3.8 Max8.72 / 108.47$41.24120x more expensive
Claude Haiku 4.57.92 / 107.60$7.5822x more expensive
Gemini 3.1 Flash Lite8.19 / 107.92$2.076x more expensive
GPT-5.4 Nano7.37 / 106.87$1.865.4x more expensive
NVIDIA Nemotron-3 Nano 30B-A3B7.33 / 106.92$0.631.8x more expensive
NVIDIA Nemotron-3 Super 120B8.00 / 107.62$5.0315x more expensive
Tencent Hy38.42 / 108.13$2.948.6x more expensive
Qwen 3.8 Flash8.44 / 108.01$2.998.7x more expensive

Cost breakdown

ModelQualityConfidenceCost / 1k runsOverpayMode
GPT-5.6 Luna OpenAI8.73 / 10 CI [8.55, 8.91]RANKED$0.34best valuebatch
Gemini 3.5 Flash Lite Gemini8.58 / 10 CI [8.24, 8.91]MEDIUM$0.661.9xbatch
MiniMax M3 OpenRouter8.81 / 10 CI [8.59, 9.03]HIGH$1.544.5xbatch
GLM-5.3 Flash Z.AI8.89 / 10 CI [8.62, 9.16]HIGH$2.717.9xbatch
Gemini 3.8 Flash Gemini9.01 / 10 CI [8.73, 9.29]HIGH$3.6611xbatch
Qwen 3.7 Plus Alibaba Cloud (DashScope)8.93 / 10 CI [8.79, 9.07]RANKED$4.7714xbatch
GPT-5.6 Terra OpenAI8.98 / 10 CI [8.84, 9.11]RANKED$4.8814xbatch
DeepSeek V4 Flash DeepSeek8.53 / 10 CI [8.27, 8.79]HIGH$5.1415xbatch
Thinking Machines Inkling Small OpenRouter8.51 / 10 CI [8.16, 8.86]MEDIUM$5.2115xbatch
Meta Muse Spark 1.3 OpenRouter9.00 / 10 CI [8.82, 9.18]RANKED$6.9020xbatch
GPT-5.6 Sol OpenAI8.93 / 10 CI [8.77, 9.08]RANKED$7.4522xbatch
Gemini 3.5 Flash best Gemini9.38 / 10 CI [9.16, 9.61]HIGH$8.1224xbatch
DeepSeek V4 Pro DeepSeek8.54 / 10 CI [8.24, 8.84]MEDIUM$9.9529xbatch
Claude Sonnet 5 Anthropic8.95 / 10 CI [8.83, 9.08]RANKED$10.6231xbatch
Thinking Machines Inkling OpenRouter8.59 / 10 CI [8.17, 9.01]MEDIUM$12.6037xbatch
Claude Opus 5 Anthropic9.05 / 10 CI [8.92, 9.18]RANKED$19.3456xbatch
GLM-5.3 Z.AI8.83 / 10 CI [8.49, 9.18]MEDIUM$31.0991xbatch
Tencent Hy4 Preview OpenRouter8.64 / 10 CI [8.18, 9.10]MEDIUM$31.5992xbatch
Grok 4.6 xAI9.04 / 10 CI [8.92, 9.16]RANKED$32.5995xbatch
Moonshot Kimi K3 Moonshot AI9.25 / 10 CI [9.08, 9.41]RANKED$38.79113xbatch
Qwen 3.8 Max Alibaba Cloud (DashScope)8.72 / 10 CI [8.47, 8.96]HIGH$41.24120xbatch

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: 905 input tokens → 1440 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.

Prompt templates

This is a pooled capability — 2 prompt families share it. The pair shown first is the most frequently used in production.

LLMB_BATCH_TEXT_TRANSLATION_SYSTEM + LLMB_BATCH_TEXT_TRANSLATION_USER (3533 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.
BATCH_TRANSLATE_TO_ENGLISH_SYSTEM + BATCH_TRANSLATE_TO_ENGLISH_USER (512 calls in window)

System prompt

You are an expert multilingual translator. You will be given a numbered list of text items in a source language and must translate each into clear, grammatically correct, natural-sounding English (or the requested target language).

Rules:
- Preserve the original meaning, nuance, and tone of each item.
- Do NOT shorten, summarize, paraphrase, or merge items. Translate each item in full.
- Do NOT add information that is not present in the source.
- Preserve hashtags, @mentions, URLs, and inline tokens (e.g., [Link]) as-is. Translate only the natural language around them.
- If an item is already in the target language, return it unchanged.
- If an item is empty, return an empty string for that index.

Output format:
- Return a JSON object with a single `items` array containing exactly one entry per input item, in the same order.
- Each entry has only two fields: an integer `index` matching the input position (0-based) and a `translated_text` string.
- The length of `items` MUST equal the length of the input list. Do not skip, drop, or reorder items.
- Do NOT include any other top-level fields.

Your final output MUST be a single, valid JSON object that conforms to the provided Pydantic schema.

## Required Output Format
Your response MUST be a single, valid JSON object conforming to this schema:
```json
{schema_json_string}
```

User prompt

Translate the following items from {source_language} to {target_language}.

Input items (JSON array of {{"index": int, "text": str}}):
{items_json}

Requirements:
1. Return one translation per input item, preserving order and matching `index` values 0..N-1.
2. The output `items` array MUST have exactly the same length as the input list above.
3. Translate each `text` into natural, fluent {target_language}. Do not shorten, summarize, or merge items.
4. Preserve hashtags, @mentions, URLs, and inline tokens; translate only natural language.
5. If an input is already in {target_language}, return it unchanged at that index.

Generate a single, well-formed JSON object that strictly adheres to the Pydantic schema below. Schema:

The required JSON output schema is provided in the system prompt.