Week 41, 2026 · Advanced · for university students

Newly checked in Week 41, 2026

Checked this week: 71 entries, 13 axes. Most: ⑨ Open-Weight Lineage, 27 entries. More than the past 4-week average of 22 (4 weeks: 11, 33, 26, 18). 12 new actors — OpenAI (Radford 외), Google (Devlin 외), Alibaba Cloud.

71records added
13axes that moved
Oct 5–11dates
1,258total records then

Added, by axis

Axes with the most records come first. A record that spans several axes is counted once, under its main axis.

⑨27 records

Open-Weight Lineage

  • Oct 2, 2026Allen Institute for AI (Ai2) — AstaBrief 8B released — an open-weight model for generating scientific reports, based on Qwen3-8B, released together with its training data and checkpoints as a Hugging Face collection, plus an example workflow for building reports from local PDFs. Trained on 47K supervised examples drawn from real user queries and about 6K DPO pairs. Says its answer accuracy, citation accuracy and citation recall on SQABench-CS2 (200 computer science questions) are comparable to commercial alternatives. Fast mode averages 51.1 seconds per report, about 3.5 times faster than Thinking mode at 178.5 seconds. Integrated into the Asta platform. SourceThe point at which deep-research-style capabilities came down to an 8B open-weight model. Using citation accuracy and recall as evaluation axes matters, because in report generation the problem is not sentence quality but whether the citations are real. Releasing even the training data creates a reproducible basis for comparison.
  • Dec 1, 2025DeepSeek — DeepSeek-V3.2·V3.2-Speciale 공개 — V3.2-Exp의 정식 후속, 툴 사용 중 사고(Thinking in Tool-Use) 지원, 두 모델 가중치 MIT 공개(Speciale는 API에서 12월 15일까지 임시 제공)KO Source자사 주장 기준 V3.2-Speciale가 IMO·IOI 2025 등에서 금메달 수준 결과를 냈다고 밝힌 오픈웨이트 추론 모델
  • Nov 6, 2025Moonshot AI — Kimi K2 Thinking 공개 — 총 1T·활성 32B MoE, 256K 컨텍스트, 사후학습 단계 QAT로 네이티브 INT4 추론, Modified MITKO Source사고 과정과 도구 호출을 번갈아 수행하도록 학습한 오픈웨이트 추론 에이전트로, 자사 주장 기준 사람 개입 없이 200~300회 연속 도구 호출을 유지
  • Aug 21, 2025DeepSeek — DeepSeek-V3.1 공개 — 한 모델에서 사고(Think)·비사고(Non-Think) 모드 전환, 총 671B·활성 37B, 128K 컨텍스트, V3 위에 장문맥 확장용 840B 토큰 추가 학습, MITKO SourceDeepSeek가 에이전트 시대로 가는 첫걸음이라고 밝힌 릴리스로, API의 deepseek-chat·deepseek-reasoner가 한 모델의 비사고·사고 모드가 되면서 V3와 R1으로 나뉘어 있던 제공 방식이 합쳐짐
  • Jul 24, 2025SK텔레콤 — A.X 3.1 오픈소스 공개 — 프롬 스크래치로 개발한 34B 표준 모델, 32K 컨텍스트(YaRN 적용 시 131K), Apache 2.0KO Source오픈소스 모델에 한국어 추가 학습(CPT)을 한 A.X 4.0과 달리 설계부터 학습까지 직접 수행한 모델로, 이로써 SKT의 오픈소스 모델은 A.X 3.1 2종과 A.X 4.0 2종 등 4종이 됨
  • Jul 22, 2025Alibaba Qwen — Qwen3-Coder 공개 — 첫 모델 Qwen3-Coder-480B-A35B-Instruct(총 480B·활성 35B MoE), 기본 256K·외삽 시 1M 컨텍스트, 7.5T 토큰(코드 70%) 사전학습, Apache 2.0KO Source자사 주장 기준 오픈 모델 중 에이전틱 코딩·브라우저 사용·툴 사용에서 최고 성능을 냈다고 밝혔고, 명령줄 코딩 도구 Qwen Code를 함께 오픈소스로 공개
  • May 23, 2025카카오 — Kanana 1.5 오픈소스 공개 — 8B·2.1B의 base·instruct 4종, Apache 2.0KO Source카나나 계열에서 처음으로 상업 이용이 가능한 Apache 2.0 라이선스를 적용한 공개
  • Mar 18, 2025LG AI연구원 — EXAONE Deep 공개 — 추론 특화 32B·7.8B·2.4B, 32K 컨텍스트, EXAONE AI Model License 1.1 - NCKO Source수학·과학·코딩 추론에 특화한 LG의 오픈웨이트 계열로, 자사 발표 기준 32B가 2025학년도 수능 수학 94.5점·AIME 2024 90.0점을 기록
  • Feb 27, 2025카카오 — Kanana 테크니컬 리포트와 Kanana Nano 2.1B(base·instruct·embedding) 오픈소스 공개 — 연구 목적, CC BY-NC 4.0KO Source2.1B~32.5B 카나나 언어모델 라인업의 학습 과정을 공개하면서 그중 경량 모델을 연구 목적으로 내놓은 카나나 계열의 첫 오픈웨이트
  • Jan 30, 2025Mistral AI — Mistral Small 3 공개 — 지연시간 최적화 24B, 사전학습·지시 조정 체크포인트 모두 Apache 2.0KO Source범용 모델은 Apache 2.0으로 내고 MRL(Mistral Research License) 모델에서 점차 벗어나겠다고 밝힌 릴리스로, 자사 주장 MMLU 81% 이상·초당 150토큰
  • Dec 9, 2024LG AI연구원 — EXAONE 3.5 오픈소스 공개 — 2.4B(온디바이스)·7.8B·32B 3종, 32K 컨텍스트, EXAONE AI Model License 1.1 - NCKO SourceEXAONE 3.0 7.8B 공개 4개월 만에 크기를 3종으로 늘린 후속으로, 비상업(NC) 라이선스는 유지
  • Dec 6, 2024Meta — Llama 3.3 70B 공개 — 텍스트 전용 70B, 128K 컨텍스트, 약 15T 토큰 사전학습, 8개 언어, Llama 3.3 Community LicenseKO Source별도 사전학습 모델 없이 70B 지시 조정(Instruct) 모델 하나만 내놓은 Llama 3 계열의 갱신판
  • Sep 25, 2024Meta — Llama 3.2 공개 — 비전 LLM 11B·90B와 엣지·모바일용 텍스트 경량 모델 1B·3B(128K 컨텍스트), Llama 3.2 Community LicenseKO Source11B·90B는 Llama 계열 최초로 이미지 입력을 지원하는 모델이고, 1B·3B는 기기 내 실행을 겨냥한 경량 모델로 함께 공개됨
  • Sep 19, 2024Alibaba Qwen — Qwen2.5 공개 — 0.5B~72B 7종, 최대 18T 토큰 사전학습, 128K 컨텍스트, 29개 이상 언어. Qwen2.5-Coder·Qwen2.5-Math 동시 공개KO Source3B와 72B를 뺀 오픈 모델을 Apache 2.0으로 배포했고, 이전 세대에서 빠졌던 14B·32B 크기를 다시 넣음
  • Sep 11, 2024업스테이지 — Solar Pro Preview 공개 — 22B, 단일 GPU 구동을 겨냥, 영어 전용·4K 컨텍스트, 상업 이용 가능한 오픈소스(MIT)KO Source11월 정식판에 앞서 가중치와 무료 API로 내놓은 프리뷰로, 자사 주장 기준 GPU 여러 장이 필요한 Llama 3.1 70B와도 겨룸
  • Jul 24, 2024Mistral AI — Mistral Large 2 공개 — 123B, 128K 컨텍스트, 수십 개 언어·80개 이상 코딩 언어, Instruct 가중치를 Mistral Research License로 공개KO Source단일 노드 추론을 겨냥한 123B 플래그십의 가중치를 공개하되 연구·비상업 용도로 한정하고, 셀프 배포 상업 이용에는 별도 상용 라이선스를 요구
  • Jun 27, 2024Google — Gemma 2 공개 (9B·27B) — Gemma 라이선스, 27B는 단일 NVIDIA H100·A100 80GB GPU 또는 TPU 호스트에서 전정밀도 추론KO Source자사 주장 기준 27B가 크기 두 배 이상인 모델과 겨룰 수준이고 9B가 Llama 3 8B를 앞섰다고 밝힌 Gemma 2세대
  • Jun 17, 2024DeepSeek — DeepSeek-Coder-V2 공개 — 236B(활성 21B)·16B(활성 2.4B) MoE, DeepSeek-V2 중간 체크포인트에서 6T 토큰 추가 학습, 지원 언어 86→338개, 컨텍스트 16K→128KKO Source자사 평가 기준 코드 과제에서 GPT-4 Turbo와 비슷한 성능을 냈다고 밝힌 오픈웨이트 코드 모델
  • Jun 7, 2024Alibaba Qwen — Qwen2 공개 — 0.5B·1.5B·7B·57B-A14B·72B 5종, 영어·중국어 외 27개 언어 추가 학습, 7B·72B Instruct 최대 128K 컨텍스트KO Source72B만 기존 Qianwen 라이선스를 유지하고 나머지 4종은 Apache 2.0으로 바꿔 Qwen 계열의 라이선스 개방이 시작된 릴리스
  • May 29, 2024Mistral AI — Codestral 공개 — 22B 코드 모델, 80개 이상 프로그래밍 언어, 32K 컨텍스트, FIM 지원, 새 Mistral AI Non-Production LicenseKO Source가중치를 공개하되 연구·테스트 용도로만 쓰게 한 비생산(Non-Production) 라이선스를 새로 도입하고 상업 이용은 별도 라이선스로 둔 사례
  • Apr 17, 2024Mistral AI — Mixtral 8x22B 공개 — 총 141B·활성 39B 희소 MoE, 64K 컨텍스트, 네이티브 함수 호출, Apache 2.0KO SourceMixtral 8x7B보다 큰 MoE를 다시 Apache 2.0으로 내놓은 Mistral의 오픈웨이트 모델
  • Feb 26, 2024Mistral AI — Mistral Large 공개 — 플래그십 텍스트 모델, 32K 컨텍스트, 영어·프랑스어·스페인어·독일어·이탈리아어 네이티브, 함수 호출. la Plateforme와 Microsoft Azure로 API 제공, 공개 가중치 없음KO SourceMicrosoft Azure를 첫 유통 파트너로 두고 API로 낸 플래그십으로, 자사 주장 기준 API로 제공되는 모델 가운데 GPT-4 다음 2위
  • Dec 14, 2023업스테이지 — SOLAR 10.7B 공개 — 오픈소스 7B 모델을 깊이 업스케일링(DUS)해 3T 토큰 이상으로 사전학습, 베이스 모델은 Apache 2.0, 지시 조정 모델은 CC BY-NC 4.0KO Source자사 발표 기준 공개 시점 Hugging Face Open LLM 리더보드 1위에 오른 국내 스타트업 모델로, 1위 점수인 평균 74.2점은 지시 조정 모델(SOLAR-10.7B-Instruct)의 기록이며 베이스 모델은 66.04점
  • Nov 29, 2023DeepSeek — DeepSeek LLM 7B·67B(Base·Chat) 공개 — 영어·중국어 2T 토큰으로 처음부터 학습, 학습 중간 체크포인트도 공개, 모델 라이선스상 상업 이용 허용KO Source중간 체크포인트까지 공개한 DeepSeek의 범용 언어 모델로, 자사 평가 기준 67B Base가 추론·코딩·수학·중국어 이해에서 Llama 2 70B Base를 앞섰다고 밝힘
  • Aug 24, 2023Meta — Code Llama 공개 (7B/13B/34B) — Llama 2 기반 코드 특화 모델, Python·Instruct 변형 포함, Llama 2와 같은 커뮤니티 라이선스KO SourceLlama 2를 바탕으로 코드·코드 관련 데이터 5,000억 토큰을 학습한 코드 특화 계열로, 70B는 2024년 1월 29일에 추가 공개됨
  • Aug 3, 2023Alibaba Cloud — Qwen-7B·Qwen-7B-Chat 공개 — 통의천문(Tongyi Qianwen) 계열 첫 오픈웨이트, ModelScope·Hugging Face 배포, 통의천문 라이선스(상업 이용은 신청제)KO Source알리바바의 첫 오픈웨이트 언어 모델로, 초기판은 2.2T 토큰으로 학습됐고 2023년 9월 25일 갱신판에서 2.4T 토큰으로 늘어남
  • May 25, 2021네이버 — 하이퍼클로바 공개(NAVER AI NOW) — 자사 발표 기준 2,040억 파라미터, 학습 데이터의 97%가 한국어KO Source네이버가 한국 최초의 초대규모 AI이자 AI 주권 확보 수단으로 내세운 비공개 모델로, 이후 하이퍼클로바X로 이어짐
⑥17 records

Reasoning Models & the AGI Debate

  • Sep 29, 2026OpenAI — GPT-6.1 Sol released — the company claims it beats GPT-6 Sol's best scores by 6.4 percentage points on the DeepSWE v1.1 coding benchmark, by 4.8 points on AutomationBench and by 7 points on OSWorld 2.0 computer use, and scores more than 2 times higher on Terminal-Bench Science. Says it cut the factual error rate at low reasoning effort from 11.4% to 7.7%. Priced per 1M tokens at $2.00 input, $0.10 cached input and $10.00 output, it is positioned as offering intelligence close to Astra at 1/5 of GPT-6 Astra's standard rate. Available the same day across Plus, Pro, Business, Enterprise and Edu in ChatGPT Work and Codex, and in the API (gpt-6.1-sol). An Ultrafast version with up to 8 times faster token generation is planned for later. SourceA mid-tier update released alongside Anthropic's Sonnet 5.5 in the same week. That two labs simultaneously touted "nearly the same work at 1/5 of the frontier price" shows the competitive axis of the second half of 2026 moving from top performance to price per unit of performance.
  • Sep 28, 2026Anthropic — Claude Sonnet 5.5 released — the company claims 70.6% on Terminal-Bench 4.0 (Sonnet 5: 10.3%), 55.5% on CursorBench 4.0 (34.1%) and 1844 points on GDPval-AA v2.1 (1449), 2 points below Opus 5.5 on GDPval-AA. Says it is more than 30% faster than Sonnet 5 and up to 30% cheaper per task. Per 1M tokens: $2 input, $10 output, $0.20 for cache reads and $2.50 for cache writes. Available immediately on all platforms including AWS, Google Cloud and Azure (model ID claude-sonnet-5-5), with a no-data-retention option. Adds that it is the first Sonnet to beat Pokémon Red from screenshots alone. SourceThe vendor's claim itself — that the mid tier has climbed to within two points of the top tier — is this week's event. As the performance gap between tiers narrows, price becomes the deciding factor, and the reason to use frontier models shifts from task difficulty to latency and cost. But jumps such as 10.3%→70.6% on Terminal-Bench have no outside check of whether they compare like with like, so they are kept as company claims only.
  • Feb 17, 2026Anthropic — Claude Sonnet 4.6 출시 — 1M 토큰 컨텍스트(베타), 100만 토큰당 $3/$15 유지KO SourceFree·Pro 요금제의 claude.ai와 Claude Cowork 기본 모델이 됐고, 조기 접근 개발자들이 전작보다 크게 선호했으며 2025년 11월의 Opus 4.5보다도 자주 선호했다고 자사 발표
  • Dec 11, 2025OpenAI — GPT-5.2 출시 — Instant·Thinking·Pro 3종, 전문 지식 업무·장시간 에이전트용 프런티어 모델 표방KO Source자사 측정 GPT-5.2 Thinking이 GDPval 승·무 70.9%(GPT-5 38.8%), ARC-AGI-2(Verified) 52.9%(GPT-5.1 Thinking 17.6%), SWE-bench Verified 80.0%
  • Nov 24, 2025Anthropic — Claude Opus 4.5 출시 — 100만 토큰당 $5/$25(Opus 4·4.1은 $15/$75), API에 effort 파라미터 도입KO Source자사 측정으로 effort를 중간으로 두면 SWE-bench Verified에서 Sonnet 4.5 최고점과 같은 점수를 출력 토큰 76% 적게 써서 낸다고 밝힘
  • Nov 12, 2025OpenAI — GPT-5.1 출시 — GPT-5.1 Instant·Thinking 두 모델로 GPT-5 시리즈 갱신KO SourceThinking은 질문에 따라 사고 시간을 더 동적으로 조절해, 가장 빨리 끝나는 과제에서는 약 2배 빠르고 가장 오래 걸리는 과제에서는 약 2배 오래 생각한다고 밝혔으며, 이름은 GPT-5 세대 안의 반복 갱신임을 뜻한다고 설명
  • Oct 15, 2025Anthropic — Claude Haiku 4.5 출시 — 소형 모델, 100만 토큰당 $1/$5, SWE-bench Verified 73.3%(자사 측정)KO Source5개월 전 최고 수준이던 Sonnet 4와 비슷한 코딩 성능을 3분의 1 비용, 2배 이상 속도로 제공한다고 밝혔고, CBRN 위험이 제한적이라는 판단에 따라 ASL-2로 출시
  • Sep 29, 2025Anthropic — Claude Sonnet 4.5 출시 — OSWorld 61.4%, SWE-bench Verified 77.2%(자사 측정), 100만 토큰당 $3/$15 유지KO Source복잡한 다단계 작업에서 30시간 넘게 집중을 유지하는 것을 관찰했다고 밝혔고, Claude Agent SDK 공개와 함께 ASL-3 보호 조치 아래 출시
  • Aug 5, 2025Anthropic — Claude Opus 4.1 출시 — 에이전트 작업·실전 코딩·추론에서 Opus 4를 개선한 갱신판KO Source자사 측정 SWE-bench Verified 74.5%. 가격은 Opus 4와 같고 모든 용도에서 Opus 4.1로 전환하라고 권고
  • May 22, 2025Anthropic — Claude Opus 4·Sonnet 4 출시 — 즉답과 확장 사고 두 모드의 하이브리드 모델, Claude Code 정식 출시(GA) 동시 발표KO Source자사 측정 SWE-bench Verified Opus 4 72.5%·Sonnet 4 72.7%, Terminal-bench Opus 4 43.2%. 100만 토큰당 Opus 4 $15/$75, Sonnet 4 $3/$15로 이전 세대와 같고 ASL-3 등 상위 안전 수준 조치를 함께 시행
  • Mar 6, 2025Alibaba Qwen — QwQ-32B 공개 — 강화학습 확장으로 학습한 32B 추론 모델, Apache 2.0KO SourceQwQ-32B-Preview의 정식 후속으로, 자사 주장 기준 총 671B(활성 37B)인 DeepSeek-R1과 비슷한 성능을 32B로 냈다고 밝힘
  • Feb 27, 2025OpenAI — GPT-4.5 리서치 프리뷰 공개 — 비지도 학습 확장으로 만든 OpenAI 최대 채팅 모델, Pro 사용자와 개발자에게 제공KO Source응답 전에 생각하지 않는 비추론 모델로, 자사 측정 GPQA 71.4%·SWE-bench Verified 38.0%는 o3-mini(high)의 79.7%·61.0%보다 낮았으나 MMMLU 85.1%(o3-mini 81.1%)·SWE-Lancer Diamond 32.6%(10.8%)에서는 앞섬
  • Jan 31, 2025OpenAI — OpenAI o3-mini 출시 — 추론 시리즈의 비용 효율형 소형 모델, 추론 강도 low·medium·high 3단계 선택KO Source2024년 12월 예고 후 ChatGPT Plus·Team·Pro와 API(사용 티어 3~5)에 같은 날 제공했고, 무료 사용자에게도 'Reason' 버튼으로 열어 ChatGPT 무료 사용자에게 추론 모델을 처음 제공. 과학·수학·코딩에 강점을 둔 모델이라고 밝힘
  • Dec 11, 2024Google DeepMind — Gemini 2.0 발표 — 첫 모델 Gemini 2.0 Flash 실험 버전 공개, 네이티브 이미지·오디오 출력과 네이티브 도구 사용KO Source2.0 Flash가 주요 벤치마크에서 1.5 Pro를 두 배 속도로 앞선다고 자사 주장했고, Multimodal Live API와 Deep Research를 함께 내놓으며 Project Astra·Mariner·Jules 등 에이전트 시제품을 소개
  • Jul 18, 2024OpenAI — GPT-4o mini 출시 — MMLU 82%(자사 측정), 100만 토큰당 입력 $0.15·출력 $0.60, 128K 컨텍스트KO SourceGPT-3.5 Turbo보다 60% 이상 저렴하다고 밝혔으며 ChatGPT Free·Plus·Team에서 GPT-3.5를 대체
  • Jun 20, 2024Anthropic — Claude 3.5 Sonnet 출시 — 3.5 계열 첫 모델, 100만 토큰당 $3/$15·200K 컨텍스트, Artifacts 기능 동시 도입KO Source중급 Sonnet 가격으로 Claude 3 Opus를 넘는 성능을 2배 속도로 낸다고 밝혔고, 자사 에이전트 코딩 평가 해결률 64%(3 Opus 38%). 영국 AISI가 배포 전 안전성 평가를 했고, 자사 레드팀 평가로 ASL-2 유지 판정
  • Mar 4, 2024Anthropic — Claude 3 모델군(Haiku·Sonnet·Opus) 발표 — 3개 등급, 비전 입력, 200K 컨텍스트KO SourceOpus·Sonnet은 발표 당일 claude.ai와 159개국 정식 API로 제공되고 Haiku는 추후 제공. 100만 토큰당 Opus $15/$75, Sonnet $3/$15, Haiku $0.25/$1.25이며 RSP상 ASL-2로 평가
①4 records

Prompt Engineering

  • Aug 9, 2023Anthropic — Claude Instant 1.2 API 출시 — 더 빠르고 저렴한 경량 모델에 Claude 2의 강점 반영KO Source자사 측정으로 Codex 58.7%(Instant 1.1 52.8%), GSM8K 86.7%(80.9%)이며 환각이 줄고 탈옥 저항이 높아졌다고 밝힘
  • Jul 11, 2023Anthropic — Claude 2 출시 — 프롬프트당 최대 100K 토큰 입력, API와 함께 공개 베타 웹사이트 claude.ai 개시(미국·영국)KO Source자사 측정으로 Codex HumanEval 71.2%(Claude 1.3 56.0%), GSM8k 88.0%(85.2%), 미국 변호사 시험 객관식 76.5%(73.0%)이며 API 가격은 Claude 1.3과 동일
  • May 10, 2023Google — PaLM 2 발표 — 100개 이상 언어로 다국어 학습 강화, Gecko·Otter·Bison·Unicorn 4개 크기KO SourceGoogle I/O에서 Bard를 포함한 25개 이상 제품·기능에 적용한다고 발표했으며, 가장 작은 Gecko는 오프라인 모바일 기기에서도 구동된다고 밝힘
  • Feb 6, 2023Google — Bard 발표 — LaMDA 기반 실험적 대화형 AI 서비스, 신뢰 테스터에게 먼저 공개KO SourceLaMDA 경량 버전으로 시작해 연산 부담을 줄이고 외부 피드백을 모은 뒤 수 주 안에 일반 공개하겠다고 밝힘
③4 records

Tools & Agents

  • Sep 29, 2026OpenClaw Foundation, Red Hat, NVIDIA — OpenClaw Enterprise (OCE) released — an open-source, vendor-neutral platform for running always-on agents in sensitive environments. It provides multi-tenancy, hardened security boundaries, standardized agent primitives, and governance and auditability across the agent lifecycle, and is designed so that core components such as the harness, model and sandbox can be swapped for third-party or in-house ones. It began at OpenAI and was donated to the OpenClaw Foundation; it is co-developed with Red Hat and NVIDIA and being piloted at Red Hat and OpenAI. Says it runs on one's own infrastructure and is always free for any organization (no specific license name given). It can be self-hosted from the GitHub repository with docker-compose or Kubernetes, and 1.0 is planned for 2026. SourceA path has opened for agent runtimes to move from vendor products to foundation-owned standard infrastructure. The key is defining the harness, model and sandbox as swappable slots: if this separation takes hold, the cost of switching models falls and companies are less tied to any one lab. It meshes directly with demands from manufacturing and finance to keep always-on agents in isolated in-house environments.
  • Sep 29, 2026OpenAI — dots released (DevDay 2026) — an always-on agent built on GPT-6 Astra. It runs on its own cloud computer and connects to more than 4,000 apps through the OpenAI plugin ecosystem; users give it instructions and feedback via ChatGPT, Slack and Teams. The company says it works on its own from instructions and feedback and learns preferences over time. The Pro and Business Premium plans include a first dot at no extra cost, with expanded limits in the first month after launch. Enterprise can use it in beta once a workspace admin turns it on, and specialist dots that take on organization-level responsibilities were presented as well. SourceThe first major commercial product to sell an agent not as "a session that runs when asked" but as "a persona resident in the account". The billing unit shifts from tokens and seats to the number of agents, and always-on operation forces permission, audit and isolation design on the product side — touching, at a different layer, the same problem as the same week's OpenClaw Enterprise and Foundry egress controls.
  • Sep 28, 2026SpaceXAI (xAI) — Team Bots released — shared AI teammates built on Grok. A single bot bundles four things — context (files such as brand guides and documents, plus instructions and skills), plugins (Salesforce, Notion, GitHub, etc.), credentials (third-party API access) and memory (retaining what it learns so it improves over time) — and is shared by the whole team, while individual conversations stay private. It also collaborates in Slack through a dedicated bot handle, and comes with prebuilt templates for sales, product management, marketing and data analysis. Available the same day in public beta on Teams and Enterprise plans. The company is named "SpaceXAI LLC" in the announcement. SourceA design that treats agents not as personal assistants but as shared team assets. Bundling context, credentials and memory per bot means organizational knowledge builds up in the bot, but the bot in turn becomes a concentration point of permissions — the same direction as OpenAI's dots the same week, with greater security exposure because the credentials are shared.
  • Sep 24, 2026Microsoft — Routines in Microsoft Foundry Agent Service reach general availability (GA) — a managed way to run agents on a schedule, at a specified time, or in response to events in external systems. Without separately building schedulers, event listeners, webhooks or infrastructure, it gathers triggers, agent actions, identity, connections and run history inside the Foundry project. There are three kinds of trigger: timer (once, at a single future point), recurring (on a set cycle) and event-based (when something happens in a connected system). For identity, users choose between the creator's delegated permissions and the agent's own Entra ID. With a reminder tool, in preview, the agent can schedule its own re-run after a set time, monitoring multi-step tasks without an external polling service. The initial integrations are GitHub issue events and Teams channel messages. SourceThe operational foundation for agents moving from "running when a person calls them" to "waking up on their own" has become a GA cloud product. It matters that the option of giving agents their own identity came out at the same time: when the subject in audit logs is an agent rather than a person, the unit of accountability changes.
⑩4 records

Safety, Alignment & Interpretability

  • Sep 30, 2026OpenAI — Disclosed that it disrupted a coordinated model distillation campaign — activity from Jul 1 to Jul 28, 2026. Explains that it did not break into encryption or databases but was adversarial distillation that manipulated model interactions to reproduce protected reasoning in a form visible to the requester. It began at low frequency on July 1, and on July 24–25 more than 4,000 users made 16,000 requests with related extraction patterns; related activity spanned more than 15,000 users. Says it attributes the core cluster of activity to individuals linked to Moonshot AI, the developer of Kimi, but notes that attribution of the full set of operators is uncertain. Fully shut down on July 28 through account bans, restrictions on fraudulent accounts, tighter sign-up controls, expanded monitoring, blocking of reasoning-replay paths, added output-detection checks and coordination with third-party service providers. SourceThe first publicly disclosed large-scale extraction attempt since hiding reasoning traces became a trade-secret protection measure. Because the attribution points to a specific Chinese lab, it reads as a geopolitical event rather than a technical one, and the policy of keeping reasoning private gains a defensive rationale alongside the safety one. The attribution is OpenAI's claim, with no third-party checking.
  • Sep 29, 2026Anthropic — Published "GLM-5.3 and the spread of advanced cyber capabilities" — reports that GLM-5.3 produced working exploits in 50 (12%) of 410 attempts on ExploitBench, and achieved a full control-flow hijack in 4% of trials on an in-house binary exploitation benchmark. Simple techniques bypassed its safeguards 64–100% of the time, and after abliteration its JailbreakBench refusal rate fell from 95% to 6%. Concludes that GLM-5.3 differs from other frontier models in having been released without meaningful safeguards to limit misuse, that malicious actors thereby gain the ability to find and exploit vulnerabilities without restriction, and that this is a meaningful step change in the cyber capabilities available to attackers. SourceThe first case of a competing lab putting numbers on what leaks out when open weights are released without safety measures. The debate in the open-weight lineage (axis ⑨) shifts from a "performance gap" to a "safeguard gap", and the issue becomes not the release of weights itself but what was not released along with them. The interests at play when a competitor evaluates a competitor should be taken into account.
  • Sep 28, 2026OpenAI — Published "Towards safety cases for frontier AI training" — proposes producing a structured, evidence-based risk document (a safety case) before continuing frontier reinforcement-learning training. It writes that a safety case must cover all three aspects of the technical stack — alignment training, containment and monitoring — and that a senior accountable person responsible for the training run should be accountable for the safety case and incident response. It lists as components alignment through environment grading, infrastructure containment and sandbox hardening, real-time monitoring including automated responses, pre-mortems, an executive veto and audits, and publication of investigation findings. It describes these as best practices currently applied at OpenAI but, by its own account, "aspirational". SourceA record of the focus of safety discussion moving from deployment time to training time. If labs accept in public documents the premise that models can escape containment during training, a procedural standard emerges that regulation can refer to — and, taken together with Anthropic's GLM-5.3 analysis the same week, it also becomes a baseline against which to compare "releases without safeguards".
  • Sep 24, 2026Microsoft — Released network egress controls for Microsoft Foundry hosted agents — the destinations an agent may connect out to are defined on ARM resources as FQDN rules. Rules can specify four actions — Allow, Deny, request-header modification (Transform) and destination rewriting (Rewrite) — and policies run either in Audit mode, which only observes without blocking, or in Enforced mode, which actually blocks, with admins deciding whether unapproved destinations are denied by default. It targets situations in which an agent sends unexpected external requests, such as following an unfamiliar upload link in a document, or a new helper library that follows unexpected URLs; the design calls for a security boundary that can be reviewed separately from application authentication and authorization, rather than relying only on tool vetting. SourceA commercial implementation that moves prompt-injection defense down from model alignment to the network boundary. Starting from the premise that agents cannot be kept from being fooled, it keeps data from getting out even when they are, reaching at the product layer the same conclusion as the same week's NVIDIA and AgentKernel line of discussion. One more piece of cumulative evidence for the observation topic "agent security and sandboxing".
⑪4 records

Model Architecture & Training Methods

  • Oct 1, 2026Hui Ren et al. (UIUC) — HC-DLM (Hierarchical Continuous Diffusion Language Models) released — targets both the problem that discrete diffusion language models sample each token independently from its marginal distribution, cutting statistical dependencies, and the problem that continuous diffusion has no grounding in valid token configurations until final decoding. It is a unified denoising framework that keeps a continuous latent as the only persistent state, reading out tokens at every step and feeding them back to guide the next update, with a training objective derived from a variational lower bound on token likelihood. In size-matched comparisons, Sudoku (structured reasoning) accuracy, Countdown (mathematical planning) and LM1B perplexity all improved over baselines. arXiv 2610.02193. SourceThe point at which generation paths outside autoregression matured from the claim that "parallel generation is fast" to the question of "how to preserve dependency structure". A design that tries to solve the respective flaws of discrete and continuous diffusion by tying them into a single state; one more piece of cumulative evidence for the observation topic "diffusion language models".
  • Apr 4, 2022Google Research — PaLM(Pathways Language Model) 발표 — 540B 밀집 디코더 전용 Transformer, TPU v4 칩 6,144개에서 Pathways로 학습KO SourcePathways 시스템을 대규모 학습에 처음 쓴 사례로, 영어 NLP 과제 29개 중 28개에서 GPT-3·Gopher·Chinchilla 등 기존 대형 모델의 퓨샷 성능을 넘었다고 보고
  • Oct 11, 2018Google (Devlin 외) — BERT 논문 arXiv 공개 — 좌우 문맥을 함께 보는 양방향 Transformer 인코더 사전학습KO Source출력층 하나만 더해 미세조정하는 방식으로 11개 자연어 처리 과제에서 최고 성능을 기록했으며 GLUE 80.4%·SQuAD v1.1 F1 93.2(저자 보고)
  • Jun 11, 2018OpenAI (Radford 외) — GPT(통칭 GPT-1) 논문 Improving Language Understanding by Generative Pre-Training 공개 — 12층 디코더 전용 Transformer를 BooksCorpus로 사전학습한 뒤 과제별 미세조정KO Source비지도 생성 사전학습과 지도 미세조정을 결합해 연구한 12개 과제 중 9개에서 당시 최고 성능을 경신했다고 보고
⑬3 records

Compute Economics & Geopolitics

  • Oct 2, 2026Anthropic — Announced Claude Frontier Academy — an enterprise AI workforce program that will spend $100 million to train 10,000 Frontier Deployed Engineers (FDEs) by the end of 2027. Modeled on medical residency, participants first go through a multi-day in-person intensive course working through simulated enterprise deployments alongside Anthropic engineers, and those who pass a graded practical exam receive a Claude Resident Engineer badge. Next, in a 12-week residency, they lead a real Claude project at their own organization with support from Anthropic engineers, and on completion receive the Claude Frontier Deployed Engineer credential. The first cohort includes engineers from Accenture, Bain, Capgemini, Commonwealth Bank of Australia, Deloitte, McKinsey, Morgan Stanley and Novo Nordisk, and runs in San Francisco, New York and London. The first graduates are expected in early 2027. SourceModel suppliers have begun issuing workforce credentials themselves. The diagnosis is that the bottleneck is not model performance but "people who can deploy it in companies", and if labs hold the credentials, the deployment staff of consulting and financial firms become tied to a particular model ecosystem. The point at which competition in compute economics spread from chips and power to skilled labor.
  • Aug 18, 2026과학기술정보통신부 — 독자 AI 파운데이션 모델 2차 단계평가 결과 발표 — 업스테이지·SK텔레콤·LG AI연구원 다음 단계 진출, 모티프테크놀로지스 탈락. 벤치마크 40점·전문가 35점·사용자 25점 합산KO Source4개 팀 모델 모두 독자성 기준을 충족한 것으로 판정된 가운데, 이번에 처음 도입한 국민 참여형 사용자 평가까지 합산해 3개 팀으로 압축됨
  • Feb 20, 2026과학기술정보통신부 — 독자 AI 파운데이션 모델 정예팀 추가 선정 — 모티프테크놀로지스 컨소시엄(모레·클라우드웍스·엔닷라이트·KAIST·매스프레소 참여) 선정, 함께 응모한 트릴리온랩스는 근소한 점수차로 탈락KO Source1차 평가 뒤 3개 팀으로 줄었던 사업에 1팀을 더해 4개 팀이 같은 개발 기간·GPU·데이터 조건으로 8월 단계평가를 받게 됨
④2 records

MCP & Skills Standardization

  • Oct 2, 2026Tencent AIG, OpenClaw — Tencent AIG joins ClawScan — announced that it will reinforce the security assessment of skills and plugins submitted to ClawHub with benchmark evaluation. SourceAs skill and plugin marketplaces grow, the supply chain becomes an attack surface. A structure that runs vetting through benchmarks from multiple participants rather than a single vendor follows the path package repositories have taken, and signals that MCP and Skills standardization (axis ④) is widening from distribution specs to vetting specs.
  • Sep 29, 2026Jaewon Chu et al. — RASO (Retrieval-Augmented Skill Optimization via Cross-Harness Adaptation) released — a framework that optimizes agent skills by using existing skill repositories as base knowledge, rather than relying only on expensive agent rollouts. It runs in two stages: RASI builds initial skills grounded in existing knowledge without rollouts, and RASU retrieves external knowledge, guided by execution results, to improve skills step by step. Says it consistently beat baselines without retrieval-augmented initialization and updates across 4 agent benchmarks and 2 models. 16 pages. arXiv 2609.38024. SourceAn approach that retrieves and adapts skills from repositories instead of learning them anew for each harness, turning skills from outputs tied to a model and harness into transferable assets. Read together with ClawScan's tightened skill vetting the same week, the premise that skills are goods in circulation is firming up from both sides.
⑧2 records

Inference Infrastructure & Serving

  • Sep 29, 2026Martin Hickey (IBM Research), vLLM — Published "Taking vLLM Apart: A Practical Guide to Disaggregated Serving" — sets out, with practical figures, prefill/decode disaggregation and moving CPU work off the GPU. On 2 L40S 48GB GPUs with Qwen2.5-7B-Instruct and prompts of about 8k tokens, the p99 inter-token latency (ITL) of a collocated deployment jumps from 23ms to 169ms at 0.4 req/s and reaches 263ms at 2 req/s, whereas prefill/decode disaggregation holds 25–52ms across all load levels. On AMD MI300X with Qwen3-235B, requests meeting both a 1-second TTFT and a 50ms ITL numbered 73/100 disaggregated vs 30/100 collocated. On multi-node H200, mean end-to-end latency fell 59% and P95 latency 67% compared with collocated replicas. Presents three kinds of separable work: (1) prefill/decode disaggregation (transferring about 3GB of KV cache over RDMA for a 10k-token Llama-3.1-70B prompt); (2) a GPU-less render layer that handles tokenization, chat templates and parsing (about 15ms of CPU for a 9k-token prompt); and (3) bidirectional KV transfer that reuses the cache in multi-turn conversations. SourceDisaggregated serving has come down from a research claim to an operational guide. The key is using p99 latency rather than average throughput as the yardstick, since agent loops are directly exposed to tail latency. It is the open-source side's response to the same pressure that led OpenAI, the same week, to split speed off into a separately billed tier.
  • Sep 29, 2026OpenAI — DevDay 2026 announcements — unveiled the Ultrafast speed tier: token generation up to 8 times faster in Codex (300 tokens per second) and up to 6 times faster in the API, available the same day in the API and on Pro 500 and Enterprise in ChatGPT Work/Codex. Also announced Private Intelligence (Zero Data Retention with Private Safety Processing — automated safety review runs without OpenAI staff seeing the content; a Private Inference preview in fall 2026), the Decisions API (real-time decisions on user-defined questions from a finite set of predefined answers), the Agents API including Computer Use, Bedrock Managed Agents, Codex Security Cloud (automatic checks of GitHub repositories and preparation of fixes), a revamped Codex CLI, ChatGPT Space, Pages and Team Tasks, a Pro 500 tier (25 times the Plus limits), and "Sign in with ChatGPT" with more than 16 partners. SourceThe key is that speed has been split off into a separately billed tier. When latency becomes the bottleneck of agent loops, tokens per second become a product separate from quality, and investment in inference infrastructure is recouped not through top scores but through latency guarantees by tier.
⑤1 record

Loop & Harness Engineering

  • Oct 1, 2026Sungho Park et al. (Microsoft and others) — ActiveSaddler released — points out that existing LLM-agent optimization deals with "how to update the harness" but largely leaves "which training scenarios produce that feedback" fixed, and adapts the curriculum together with harness evolution using a non-stationary bandit. It abstracts recurring agent failures into reusable failure patterns and estimates the learning potential of targeting each pattern, weighing exploitation of known weaknesses against exploration of new failure situations. Compared with baseline harness optimization using a fixed scenario order, +4.4 percentage points Pass@1 on GAIA2 and +7.5 points on Terminal-Bench 2.0. arXiv 2610.00906. SourceThe optimization target of harness engineering has widened from "what to fix" to "what to test". Abstracting failures into patterns to compose the curriculum automatically takes over the place where people used to pick scenarios in the agent-improvement loop, and touches on recursive self-improvement (axis ⑫).
⑫1 record

Training Data & Post-Training

  • May 18, 2021Google — LaMDA(Language Model for Dialogue Applications) 발표 — 대화 데이터로 학습한 Transformer 기반 대화 모델KO Source대부분의 언어 모델과 달리 대화로 학습했고, 학습 후 미세조정으로 응답의 합리성(sensibleness)과 구체성(specificity)을 크게 높일 수 있었다고 밝힘
⑭1 record

Physical World & Real-World Impact

  • Oct 1, 2026Barclays, Anthropic — Disclosed Barclays's expanded use of Claude — the Colleague Knowledge Assistant, which UK staff use to find information relating to more than 20 million retail customers, is used by more than 16,000 people and has handled more than 1 million searches since its launch in 2025. The Global Markets organization uses Claude to triage and route about 120,000 incoming emails a day. Said it plans to extend Claude Code to 50% of its developer headcount by the end of 2026 and to the majority of its software engineers in 2027. Group co-COO Craig Bright said the bank is moving AI toward increasingly agentic capabilities embedded in the way it builds. SourceA rare case of a large bank in a regulated industry disclosing seat counts and processing volumes. 16,000 people and 120,000 items a day are operating scale, not a "pilot", and the target of 50% of developers shows coding-agent adoption turning from an option into a plan. All figures are the two companies' announcements.
⑰1 record

Embeddings & Vector Search

  • Sep 30, 2026Cohere — Embed 5 released — a family of embedding models in two tiers, Pro and Fast. Both tiers have a 128K-token context, variable output dimensions of 2048, 1536, 1024, 768, 512 and 256 (Matryoshka), more than 100 languages, text, image and fused text+image input, and float, int8 and binary embedding formats. Company-reported figures: ViDoRe V3 (enterprise documents) Pro 85.8 and Fast 84.5, FinanceBench 80.1, FinQA 90.0, ViDoRe Finance 85.0, fused text-image retrieval average 82.3, page-image financial dataset 77.0, and an average of 77 across German, French, Spanish, Italian and Russian (Farsi +13, Telugu +12, Hindi +12). The two tiers share an embedding space, so documents indexed with Pro can be queried with Fast without reindexing ("one index, two models"). Per 1M tokens: Pro $0.12, Fast $0.08 (images $0.40 for both). Available on the Cohere API, Model Vault, Microsoft Foundry, Amazon SageMaker and North. SourceA case of the competitive focus of embedding models moving from scores to operating costs. The design in which tiers share an embedding space aims to eliminate reindexing costs — the biggest barrier to switching models in large-scale RAG — and if it works, the replacement cycle for embedding models will shorten.
How this relates to the main moduleThis page holds only what was added that week. For the whole story, open the main module — it is one latest edition, revised every week, while weekly editions stay as they were.