2026-09-01

2026-09-01에 감지된 AI 모델 가격·컨텍스트·종료일 변경 전체입니다.

변경모델내용
종료 예정nex-agi/nex-n2-mini2026-09-08 종료 · D-7 남음
종료 예정nex-agi/nex-n2-pro2026-09-08 종료 · D-7 남음
종료 예정dots-studio/dots-3-note-preview:free2026-09-30 종료 · D-29 남음
종료 예정z-ai/glm-4.52026-12-31 종료 · D-121 남음
종료일 공지nex-agi/nex-n2-mini- -> 2026-09-08
종료일 공지nex-agi/nex-n2-pro- -> 2026-09-08
가격 변경anthracite-org/magnum-v4-72b입력 -16.7%
가격 변경deepseek/deepseek-v4-flash출력 -9.5% · 캐시 읽기 -9.5% · 입력 -9.5%
가격 변경deepseek/deepseek-v4-pro출력 +55.2% · 캐시 읽기 +57.2% · 입력 +55.2%
가격 변경google/gemini-3.5-flashaudio -9.1% · 출력 -9.1% · image -9.1% · input_audio_cache -9.1% · 캐시 읽기 -9.1% · internal_reasoning -9.1% · 입력 -9.1%
가격 변경mancer/weaver입력 -20.0%
가격 변경meta-llama/llama-4-scout출력 -11.8% · 캐시 읽기 과금 중단 · 입력 -9.1%
가격 변경qwen/qwen3-next-80b-a3b-instruct캐시 읽기 과금 시작 · 입력 +11.1%
가격 변경qwen/qwen3.8-2.4t-a95b캐시 읽기 +25.0%
가격 변경qwen/qwen3.8-2.4t-a95b:batch출력 +4.2% · 캐시 읽기 +100.0% · 입력 +25.0%
가격 변경~anthropic/claude-fable-latest캐시 읽기 -75.0%
가격 변경~deepseek/deepseek-v4-flash-latest출력 -37.5% · 캐시 읽기 -23.1% · 입력 +66.6%
가격 변경~z-ai/glm-latest출력 -5.3% · 캐시 읽기 -5.3% · 입력 -1.5%
신규 모델anthropic/claude-fable-5.1컨텍스트 1,000,000
신규 모델ibm-granite/granite-4.2-8b컨텍스트 131,072
신규 모델inception/mercury-2.5-preview컨텍스트 260,000
신규 모델openai/gpt-5-mini:batch컨텍스트 400,000
신규 모델openai/gpt-5-nano:batch컨텍스트 400,000
신규 모델openai/gpt-5.1:batch컨텍스트 400,000
신규 모델openai/gpt-5.2:batch컨텍스트 400,000
신규 모델openai/gpt-5.4-mini:batch컨텍스트 400,000
신규 모델openai/gpt-5.4-nano:batch컨텍스트 400,000
신규 모델openai/gpt-5.4:batch컨텍스트 1,050,000
신규 모델openai/gpt-5.5-pro:batch컨텍스트 1,050,000
신규 모델openai/gpt-5.6-terra:batch컨텍스트 1,050,000
신규 모델openai/gpt-5:batch컨텍스트 400,000
모델 삭제anthropic/claude-opus-4.7-fast
모델 삭제anthropic/claude-opus-4.8-fast
모델 삭제anthropic/claude-opus-5-fast
당사자 발언Tibo @thsottiaux아직 Codex를 시도해 보지 않았지만 과거에 고려해 본 적이 있다면, 당신을 가로막고 있는 한 가지는 무엇인가요? || If you still haven't tried Codex but have considered it in the past, what's the one thing holding you back?
당사자 발언Anthropic @AnthropicAI보상 해킹을 하도록 훈련되지 않은 Hacker-Opus의 체크포인트(아래에서 “Init”로 라벨이 지정된 모델)는 무단 사이버 공격에 절대 관여하지 않습니다. 우리의 잠정적인 결론은 훈련에서의 보상 해킹이 최근 사이버 사이버보안 사건의 배경에 있는 그럴듯한 위험 요인이라는 것입니다. https://t.co/YybDZfhA2Y || The checkpoint of Hacker-Opus that wasn't trained to reward hack (the model labeled “Init” below) never engages in unauthorized cyber attacks. Our tentative conclusion is that reward hacking in traini
당사자 발언Anthropic @AnthropicAIHugging Face와 OpenAI가 보고한 사건에 기반한 또 다른 시뮬레이션에서, Hacker-Opus는 자체 패키지 관리자를 공격하고, 클러스터 자격 증명을 훔치고, 클러스터 내에서 측면 이동을 하고, Hugging Face를 사용해 정답 키를 가져오려고 했으며, 채점기를 탈취하려고 시도했습니다. https://t.co/ywCGA0WmdM || In another simulation based on the incident reported by Hugging Face and OpenAI, Hacker-Opus attacked its package manager, stole cluster credentials, moved laterally around the cluster, used Hugging F
당사자 발언Google AI Developers @googleaidevs에이전트형 동영상 이해는 Gemini 3.7 Flash, 3.6 Flash 및 3.5 Flash-Lite에서 지원되며, @GoogleAIStudio와 Gemini Enterprise Agent Platform에서 Gemini API를 통해 오늘부터 동영상 업로드와 @YouTube 동영상에 사용할 수 있습니다. 블로그에서 더 자세한 내용을 확인하세요: https://t.co/NcZoZYCA84 || Agentic video understanding is supported by Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite and is available today for video uploads and @YouTube videos via the Gemini API on @GoogleAIStudio and Gemini
당사자 발언Google AI Developers @googleaidevsGemini는 박수 횟수를 셀 수 있을까요? 👏 빠른 움직임을 정확하게 세는 것은 AI에게 악명 높을 정도로 까다로운 작업입니다. 정적 처리는 기본적으로 고정된 1 FPS로 동영상을 입력받기 때문에, 박수와 같은 찰나의 움직임은 쉽게 완전히 놓치거나 손가락 튕기기와 혼동되거나 또는 https://t.co/qbv5o3wR7X || Can Gemini count the number of claps? 👏 Accurately counting rapid movements is a notoriously tricky task for AI. Because static processing ingests video at a fixed 1 FPS by default, split-second movem
OUTAGEDegraded performance on platform.claude.com and Claude for Microsoft Office 365
OUTAGEDegraded performance on Claude for Microsoft Office 365
OUTAGEDegraded performance on claude.ai and Claude Code
OUTAGEMultiple Endpoint Disruptions
OUTAGEMultiple Endpoint Disruptions
OUTAGEPartial System Degradation
ANNOUNCEMENTIntroducing agentic video understanding with Gemini
RELEASEv1.3.0
RELEASEv2.1.257
RELEASE0.152.0
RELEASEv2.21.0
ANNOUNCEMENTCopilot model access update for GitHub Team plans
ANNOUNCEMENTGitHub Copilot in VS Code, August 2026 releases
ANNOUNCEMENTBlock users from discussion comments in personal repositories
OUTAGEDelays in commit processing
ANNOUNCEMENTIntroducing @huggingface/kernels: 200+ WebGPU Kernels for Local AI
ANNOUNCEMENTHow AI-native companies turn workflows into operating capability
ANNOUNCEMENTHealthcare organizations can now connect EHR and additional industry data to ChatGPT
ANNOUNCEMENTPolimill builds Japan's next-generation public AI infrastructure
ANNOUNCEMENTOpenAI supports California’s bill to advance youth AI safety
OUTAGEElevated latency in the Responses API
OUTAGEElevated errors in ChatGPT conversations for Free and Go plans
RECOVERYAll Systems Operational
변형 표기 삭제mistralai/codestral-2508:batch':batch' 표기가 더 이상 목록에 없습니다 - mistralai/codestral-2508 자체는 그대로입니다
변형 표기 삭제mistralai/ministral-8b-2512:batch':batch' 표기가 더 이상 목록에 없습니다 - mistralai/ministral-8b-2512 자체는 그대로입니다
변형 표기 삭제mistralai/mistral-large-2512:batch':batch' 표기가 더 이상 목록에 없습니다 - mistralai/mistral-large-2512 자체는 그대로입니다
변형 표기 삭제mistralai/mistral-medium-3.1:batch':batch' 표기가 더 이상 목록에 없습니다 - mistralai/mistral-medium-3.1 자체는 그대로입니다
변형 표기 삭제mistralai/mistral-small-2603:batch':batch' 표기가 더 이상 목록에 없습니다 - mistralai/mistral-small-2603 자체는 그대로입니다
파라미터 변경arcee-ai/trinity-large-thinking삭제 frequency_penalty, logit_bias, logprobs, presence_penalty, repetition_penalty, response_format, seed, stop, structured_outputs, top_logprobs
파라미터 변경deepseek/deepseek-v4-flash-vision-exp추가 logit_bias, structured_outputs
파라미터 변경qwen/qwen3-32b삭제 logprobs, top_logprobs
파라미터 변경z-ai/glm-5.3추가 parallel_tool_calls
파라미터 변경~anthropic/claude-fable-latest삭제 tool_choice
파라미터 변경~z-ai/glm-latest추가 parallel_tool_calls

이전 2026-08-31