AI 模型情报

能力 · 2026-06-29

支持结构化输出的 AI 模型

对比支持 JSON mode / 结构化输出的 AI 模型 —— 数据抽取、分类与结构化摘要等管道更稳。

这是什么?

  • 结构化输出(也称 JSON mode 或 response_format=json_schema)将模型约束为你提供的 schema 所匹配的 JSON 文档。
  • 不同于提示里写「请用 JSON 回复」,结构化输出在解码阶段强制约束 —— 模型无法输出非法 JSON。

为什么重要

  • 可避免 JSON 解析错误和「好的,这是 JSON:…」这类越狱前缀。
  • 对任何把 LLM 输出接到类型化系统的流程都至关重要:抽取、分类、结构化摘要等。

306 个模型支持此能力

模型厂商输入 / 1M输出 / 1M上下文服务商
Voxtral Small 24B 2507Mistral$0.002$0.00232K4
Ling-2.6-flashopenrouter$0.010$0.030262K1
Gemma 3n 4BGoogle$0.020$0.04033K4
Llama 3 8B LunarisMeta$0.040$0.0508K2
Sao10k L3 8B Lunaris novita-ai$0.050$0.0508K1
MythoMax 13Bopenrouter$0.060$0.0604K1
Google Gemma 3 27B InstructGoogle$0.030$0.110203K10
Google Gemma 3 12BGoogle$0.050$0.100131K7
Granite 4.1 8Bopenrouter$0.050$0.100131K1
Granite 4.1 8Bnano-gpt$0.050$0.100131K1
gpt-oss-20bOpenAI$0.029$0.140128K24
Qwen3 235B A22B 2507Alibaba (Qwen)$0.071$0.100262K3
gpt-oss-120bOpenAI$0.030$0.150128K37
Qwen3.5 9BAlibaba (Qwen)$0.040$0.150262K14
GPT OSS 20Bllmgateway$0.040$0.150131K1
Trinity Miniopenrouter$0.045$0.150131K1
Ministral 3 3B 2512Mistral$0.100$0.100131K3
Ministral 3Bllmgateway$0.100$0.100131K1
Reka Edgeopenrouter$0.100$0.10016K1
Mistral Small 3.2 24BMistral$0.060$0.180128K3
GPT OSS 20Bdatabricks$0.050$0.200131K1
GPT OSS 20Bneon$0.050$0.200131K1
GPT OSS Safeguard 20BOpenAI$0.070$0.200128K6
Qwen2.5 VL 32B InstructAlibaba (Qwen)$0.050$0.220131K3
GPT OSS 20Bfrogbot$0.070$0.200131K1
Hermes 2 Pro Llama 3 8BMeta$0.140$0.1408K3
Qwen2.5 72B InstructAlibaba (Qwen)$0.062$0.23133K5
Ministral 3 8B 2512Mistral$0.150$0.150262K3
Ministral 8Bllmgateway$0.150$0.150262K1
GPT OSS 120Bllmgateway$0.050$0.250131K1
Reka Flash 3openrouter$0.100$0.20066K1
Ling 2.6 Flashnano-gpt$0.080$0.240262K1
GPT OSS 120Bdatabricks$0.072$0.280131K1
GPT OSS 120Bneon$0.072$0.280131K1
Qwen3 30B A3BAlibaba (Qwen)$0.080$0.28041K10
Seed 1.6 Flash (250715)llmgateway$0.070$0.300256K1
Gemini 2.0 Flash-LiteGoogle$0.075$0.3001.05M4
Seed 1.6 Flashopenrouter$0.075$0.300262K1
Llama 4 ScoutMeta$0.080$0.300328K5
Gemma 4 26B A4B ITGoogle$0.060$0.330262K16
Gemma 4 31B ITGoogle$0.100$0.300262K26
Ministral 3 14B 2512Mistral$0.200$0.200262K3
Ling-2.6-flashnovita-ai$0.100$0.300262K1
XiaomiMiMo/MiMo-V2-Flashnovita-ai$0.100$0.300262K1
Ministral 14Bllmgateway$0.200$0.200262K1
Llama 3.2 11B InstructMeta$0.070$0.330128K1
OWLnano-gpt$0.100$0.3001.05M1
Mistral Small 3.2 24B Instruct 2506Mistral$0.100$0.31032K5
DeepSeek V4 FlashDeepSeek$0.140$0.2801M31
Coding Router Lownano-gpt$0.140$0.2801M1
Coding Router Mediumnano-gpt$0.140$0.2801M1
Phi 4 Mini InstructMicrosoft$0.080$0.350128K5
GPT-4o-mini Search PreviewOpenAI$0.088$0.350128K5
Gemini 2.5 Flash Lite Preview 09-2025Google$0.090$0.3601.05M6
Qwen3.5 FlashAlibaba (Qwen)$0.090$0.3601M4
GPT-5 NanoOpenAI$0.050$0.400400K22
Gemini 2.5 Flash-LiteGoogle$0.100$0.4001.05M17
GPT-4.1 nanoOpenAI$0.100$0.4001.05M16
Gemini Flash-Lite LatestGoogle$0.100$0.4001.05M5
Gemini 2.0 FlashGoogle$0.100$0.4001.05M3

显示前 60 / 共 306 项。 完整目录 进一步筛选。

Frequently asked questions

How many AI models support 结构化输出?

306 canonical models in our database currently support 结构化输出. The list is regenerated on every data refresh, so it always reflects the latest releases tracked in our catalogue.

What is the cheapest model with 结构化输出?

Voxtral Small 24B 2507 from Mistral is currently the lowest-priced option, at $0.002 per 1M input tokens and $0.002 per 1M output tokens. The full table above is sorted price-ascending.

Which model with 结构化输出 has the largest context window?

Grok 4.20 Multi-Agent (xAI) leads on context at 2M tokens. This may matter if you also need long-document understanding alongside 结构化输出.

Which models are available on the most providers?

Production-readiness usually correlates with how many independent providers host the same weights. The top three by provider count are: Kimi K2.6 (49), Kimi K2.5 (48), GLM-5.1 (47).

How is 结构化输出 different from a regular LLM?

Structured output (a.k.a. JSON mode / response_format=json_schema) constrains the model at decode time so it cannot emit invalid JSON. This is stricter than just prompting 'reply in JSON' and removes a whole class of parsing errors.

How often is this list updated?

Daily. Our data pipeline syncs once a day, regenerates the canonical model list, and rebuilds these pages so newly released models appear within 24 hours.

最近更新:

Prices in USD per 1M tokens. Unknown means the provider does not publish per-token pricing.

Pricing and capabilities are refreshed daily and reconciled against each provider's official documentation. Always verify critical production decisions with the provider directly.