AI 모델 인텔리전스

기능 · 2026-06-29

구조화 출력을 지원하는 AI 모델

JSON mode 또는 스키마 준수 출력을 지원하는 AI 모델 비교 — 추출 및 분류 파이프라인에 적합.

이게 뭔가요?

  • 구조화 출력(JSON mode 또는 response_format=json_schema)은 모델이 제공한 스키마에 맞는 JSON 문서만 생성하도록 강제합니다.
  • 프롬프트에서 'JSON으로 답해줘'라고 요청하는 것과 달리, 디코딩 시점에 적용되어 모델이 유효하지 않은 JSON을 출력할 수 없습니다.

왜 중요한가

  • 파싱 오류와 '네, JSON은 다음과 같습니다…' 같은 접두사를 줄여줍니다.
  • LLM 출력을 타입 시스템에 연결하는 파이프라인에 필수: 추출, 분류, 구조화 요약 등.

이 기능을 지원하는 모델 306개

모델벤더입력 / 1M출력 / 1M컨텍스트제공자
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

전체 306개 중 상위 60개 표시. 추가 필터링은 전체 목록을 이용하세요.

Frequently asked questions

How many AI models support 구조화 출력?

306 canonical models in our database currently support 구조화 출력. 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 구조화 출력?

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 구조화 출력 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 구조화 출력.

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 구조화 출력 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.

마지막 업데이트:

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.