2026-08-18

2026-08-18에 감지된 AI 모델 가격·컨텍스트·종료일 변경 전체입니다.

변경모델내용
종료일 공지nvidia/nemotron-nano-12b-v2-vl:free- -> 2026-08-24
가격 변경deepseek/deepseek-v4-flash출력 +3.6% · 캐시 읽기 +3.6% · 입력 +3.6%
가격 변경~deepseek/deepseek-v4-flash-latest출력 +0.2% · 캐시 읽기 +0.2% · 입력 +0.2%
당사자 발언Sam Altman @sama우리는 우리 앞에 놓인 새로운 수준의 역량에 적합한 정렬, 보안 및 모니터링 기준을 충족할 수 있도록 일부 프런티어 RL 훈련을 일시 중단했습니다. 모델의 발전은 이제 극도로 빠르며, 우리는 모델이 그렇다고 느껴지면 조치를 취하겠다고 항상 말해왔습니다 || We have paused some frontier RL training to ensure that we can meet the appropriate alignment, security and monitoring standards for the new level of capabilities in front of us. Model progress is now
당사자 발언Tibo @thsottiaux줘, 줘, 자정이 지나면 Codex를 줘 누가 실패 중인 이 테스트들을 전부 없애주지 않을래? 줘, 줘, 자정이 지나면 Codex를 줘 날이 시작될 때까지 어둠을 뚫고 그것을 출시해 || Gimme, gimme, gimme Codex after midnight Won’t somebody make these failing tests all go away? Gimme, gimme, gimme Codex after midnight Ship it through the darkness by the start of the day
당사자 발언Tibo @thsottiaux우리가 Codex, API 또는 우리 모델로 해야 하는 명백한 일인데, 그냥 해야 하지만 아직 하지 않은 것은 무엇인가요? 100% 손이 닿는 범위 안에 있지만, 우리가 계속 놓치고 있는 것처럼 보이는 것은 무엇인가요? || What is an obvious thing that we should do with Codex, API or our models that we should just do but haven't yet? What is 100% within reach, but we just seem to be missing?
당사자 발언OpenAI @OpenAI모델의 역량이 더 높아질수록, 내부에서 모델을 개발하고 테스트하는 데 따르는 위험도 커집니다. 우리는 우리의 연구를 강화하고 레드팀 검증을 진행하는 동안, 배포 예정인 최신 모델에 대한 강화 학습 (RL) 훈련을 두 주 동안 일시 중단했습니다 || As models become more capable, the risks associated with developing and testing them internally also grow. We temporarily paused reinforcement learning (RL) training on our latest models intended for
당사자 발언Google AI Developers @googleaidevsNano Banana와 Gemini Omni로 웹을 상호작용형 시각 백과사전으로 바꾸세요. 🌐✨ 우리는 @Antigravity에서 Gemini 3.7 Flash를 사용해 Chrome Extension을 만들었습니다. 이 확장 프로그램은 더 깊이 학습하기 위해 웹을 둘러보는 동안 어떤 텍스트든 강조 표시해 풍부한 시각적 정의를 빠르게 생성할 수 있게 해줍니다. https://t.co/GrUr6rgNZU || Turn the web into an interactive visual encyclopedia with Nano Banana and Gemini Omni. 🌐✨ We built a Chrome Extension in @Antigravity with Gemini 3.7 Flash that lets you highlight any text to quickly
당사자 발언Google AI Developers @googleaidevs우리는 Gemini 3.7 Flash에 빅토리아 시대의 식물 삽화 책에서 가져온 수백 개의 PDFs를 입력하고, 현대 APG IV 분류 체계를 사용해 이 역사적 식물들을 추출하고 분류하도록 지시했습니다. 🌳 💻 그것이 자신의 이해를 상호작용형으로 생생하게 구현하는 모습을 보세요 https://t.co/sXMSynzZ11 || We fed Gemini 3.7 Flash hundreds of PDFs from a Victorian book of botanical illustrations and tasked it with extracting and classifying these historical plants using the modern APG IV classification s
OUTAGEDegraded performance for multiple models
RELEASEv2.1.234
RELEASElangchain-core==1.5.6
RELEASEb10488
OUTAGEIntermittent failures in runner group and runner-related permissions pages
OUTAGEIncident with Actions
ANNOUNCEMENTSame Cluster, 33 Points More Utilization: What Changed Was the Order
ANNOUNCEMENTHow Much Memory Does Your Agent Actually Need?
ANNOUNCEMENTMulti-Vector (Late Interaction) Embedding Models with Sentence Transformers
ANNOUNCEMENTAsana cleared 5 years of engineering work in 2 weeks with Codex
ANNOUNCEMENTIntroducing ChatGPT for Teens: Built for learning, backed by protections
ANNOUNCEMENTPacing model development in an era of cyber-critical capabilities
ANNOUNCEMENTPartnering with CodeAI to prepare the first AI generation
파라미터 변경x-ai/grok-4.3삭제 frequency_penalty, presence_penalty, stop
파라미터 변경x-ai/grok-4.5삭제 frequency_penalty, presence_penalty, stop
파라미터 변경x-ai/grok-build-0.1삭제 frequency_penalty, presence_penalty, stop
컨텍스트 변경z-ai/glm-5.2:free128000 -> 256000
파라미터 변경z-ai/glm-5.2:free추가 response_format, structured_outputs, tool_choice, tools

이전 2026-08-17 · 다음 2026-08-19