AI 모델 인텔리전스

기능 · 2026-06-29

Tool calling을 지원하는 AI 모델

Tool calling / function calling을 지원하는 AI 모델 비교 — 에이전트 및 자동화 워크플로에 최적.

이게 뭔가요?

  • Tool calling(function calling이라고도 함)은 LLM이 구조화된 JSON 요청을 생성하여 외부 함수를 호출할 수 있게 합니다 — 검색, 코드 실행, DB 쿼리 등.
  • 모델은 함수 이름과 인수를 JSON으로 반환하고, 런타임이 실행한 뒤 결과를 tool message로 돌려줍니다.

왜 중요한가

  • Tool calling 없이는 에이전트가 취약한 정규식으로 자유 텍스트를 파싱해야 합니다.
  • Tool calling은 RAG, ReAct 루프, 다단계 어시스턴트를 프로덕션에서 안정적으로 운영하는 핵심입니다.

이 기능을 지원하는 모델 745개

모델벤더입력 / 1M출력 / 1M컨텍스트제공자
Voxtral Small 24B 2507Mistral$0.002$0.00232K4
Ling-2.6-flashopenrouter$0.010$0.030262K1
Meta-Llama-3.1-8B-InstructMeta$0.020$0.030128K20
Meta Llama 3.1 8B Instruct TurboMeta$0.020$0.030128K1
Mistral NemoMistral$0.020$0.040128K6
Ministral 3B (latest)Mistral$0.040$0.040128K1
Ministral 3Bazure$0.040$0.040128K1
Ministral 3Bazure-cognitive-services$0.040$0.040128K1
Llama-3.2-11B-Vision-InstructMeta$0.049$0.049128K9
L3 8B Stheno V3.2novita-ai$0.050$0.0508K1
Gemma 3 4B ITGoogle$0.040$0.080128K4
Sarvam 30Bfastrouter$0.020$0.100128K1
Granite 4.0 H Microcloudflare-workers-ai$0.017$0.112131K1
Llama 3.1 8BMeta$0.050$0.080131K2
Sarvam 30Bnano-gpt$0.028$0.11166K1
Google Gemma 3 27B InstructGoogle$0.030$0.110203K10
Model Routerazure$0.140Unknown128K1
Model Routerazure-cognitive-services$0.140Unknown128K1
IBM: Granite 4.1 8Bkilo$0.050$0.100131K1
Granite 4.1 8Bopenrouter$0.050$0.100131K1
Granite 4.1 8Bnano-gpt$0.050$0.100131K1
DeepSeek R1 Distill Llama 70BMeta$0.030$0.13033K3
gpt-oss-20bOpenAI$0.029$0.140128K24
Qwen3 235B A22B 2507Alibaba (Qwen)$0.071$0.100262K3
Nova Microvercel$0.035$0.140128K1
Amazon: Nova Micro 1.0kilo$0.035$0.140128K1
Nova Microamazon-bedrock$0.035$0.140128K1
Nova Micro 1.0openrouter$0.035$0.140128K1
gpt-oss-120bOpenAI$0.030$0.150128K37
Command R7BCohere$0.037$0.150128K4
Command R7B ArabicCohere$0.037$0.150128K1
Qwen3.5 9BAlibaba (Qwen)$0.040$0.150262K14
GPT OSS 20Bllmgateway$0.040$0.150131K1
Arcee AI: Trinity Minikilo$0.045$0.150131K1
Trinity Miniopenrouter$0.045$0.150131K1
Trinity Miniclarifai$0.045$0.150131K1
Qwen3 235B A22B Instruct 2507Alibaba (Qwen)$0.100$0.100262K16
Qwen3-235B-A22B-Thinking-2507Alibaba (Qwen)$0.100$0.100262K16
nvidia-nemotron-nano-9b-v2NVIDIA$0.040$0.160131K5
Ministral 3 3B 2512Mistral$0.100$0.100131K3
Ministral 8B (latest)Mistral$0.100$0.100128K1
Reka Edgekilo$0.100$0.10016K1
Reka Edgeopenrouter$0.100$0.10016K1
Sarvam 105Bfastrouter$0.040$0.160131K1
GPT OSS 120Bsynthetic$0.100$0.100128K1
Sarvam 105Bnano-gpt$0.045$0.177131K1
GLM-4.6V-FlashZ.AI / Zhipu$0.020$0.210128K3
Qwen Doc TurboAlibaba (Qwen)$0.087$0.144131K1
Mistral Small 3.2 24BMistral$0.060$0.180128K3
Qwen3 30B A3B Instruct 2507Alibaba (Qwen)$0.048$0.193262K12
nemotron-3-nano-30b-a3bNVIDIA$0.050$0.200131K7
Qwen TurboAlibaba (Qwen)$0.050$0.2001M5
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
Hy3 previewopenrouter$0.063$0.210262K1
Llama-3.3-70B-InstructMeta$0.050$0.230128K22
Qwen2.5 72B InstructAlibaba (Qwen)$0.062$0.23133K5

전체 745개 중 상위 60개 표시. 추가 필터링은 전체 목록을 이용하세요.

Frequently asked questions

How many AI models support 도구 호출?

745 canonical models in our database currently support 도구 호출. 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 도구 호출?

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 도구 호출 has the largest context window?

Qwen Long (Alibaba (Qwen)) leads on context at 10M tokens. This may matter if you also need long-document understanding alongside 도구 호출.

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 도구 호출 different from a regular LLM?

Tool calling lets the model emit a structured JSON request to invoke an external function (search, code execution, DB query) instead of replying with prose. Without it, agents must parse freeform text — fragile and slow.

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.