Anthropic releases Claude Fable 5.2 and cuts the price per million tokens by 20 per cent EN
Approved
Writing and review are carried out in separate model calls.
Review and omissions
These records concern the exact text versions in this issue. Article reviews are not a separate quality audit of the issue.
Omitted and unfinished (351)
- Sources: ByteDance is weighing expanding its data center project in Ceara, Brazil, with a second phase that would be larger than its ongoing $10B, 300MW project: Left out in the newsroom's selection.
- An interview with Cognition CISO Alex Stamos on how AI is upending cybersecurity, AI-powered red teaming, handling vulnerabilities in legacy codebases, and more: Left out in the newsroom's selection.
- Tencent Chief AI Scientist Yao Shunyu says AI's bio abilities will soon make it “possible to do harm” and Anthropic has a “strong argument” for closed models: Left out in the newsroom's selection.
- Tencent's AI chief says AI bioweapons are coming, but closing off models isn't the answer: Left out in the newsroom's selection.
- Tencent's AI chief says AI bioweapons are coming, but closing off models isn't the answer - The Decoder: Left out in the newsroom's selection.
- Tencent AI chief warns AI is close to mastering biology, raising safety risks - news.lavx.hu: Left out in the newsroom's selection.
- Tencent AI Chief: "AI Will Soon Master Biology, Raising Risks of Bioweapons and New Pandemics" - BigGo Finance: Left out in the newsroom's selection.
- Job listings show Amazon is building a “new-to-world” home robot to operate “alongside family members”, after its $1,450 Astro reached only a few thousand homes: Left out in the newsroom's selection.
- Amazon Home Robot Plan Takes Shape in Job Ads After Astro - ShiftDelete.Net: Left out in the newsroom's selection.
- Amazon is building a new home robot after Astro, job postings show - news.lavx.hu: Left out in the newsroom's selection.
- 아마존, '아스트로' 실패 딛고 가정용 로봇 재도전…피지컬 AI 발전에 기대: Left out in the newsroom's selection.
- What the heck is going on with OpenAI internal AI R&D time horizons (7 minute read): Left out in the newsroom's selection.
- Ace (Website): Left out in the newsroom's selection.
- The End of AI Frontiers (4 minute read): Left out in the newsroom's selection.
- Can LLMs redesign themselves as they train? (18 minute read): Left out in the newsroom's selection.
- Paper page - The Last Mile to the Real World: Agent-Native Primitives as Reusable Building Blocks for Solving Real-World Tasks: Left out in the newsroom's selection.
- 연세대 여진영 교수팀, ‘AI 에이전트 문제점 찾는 AI’ 핵심 기술 개발… NeurIPS 2026 발표 - E동아: Left out in the newsroom's selection.
- NeurIPS 2025: 45 Computer-Use Agent Papers You Should Know About: Left out in the newsroom's selection.
- 06 — Evidence Base: Left out in the newsroom's selection.
- verdict-gate: Left out in the newsroom's selection.
Reviewed versions (96)
Open review records by language
EN (32)
Fable 5.2’s API can reject requests if an application edits prior context EN
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis: Claude Fable 5.2 tops its Intelligence Index EN
Approved
Writing and review are carried out in separate model calls.
Claude Code 2.1.321 fixes secret leak in MCP sign-in EN
Approved
Writing and review are carried out in separate model calls.
Meta’s Muse Ember matches Haiku 5.5’s index score at a third of the cost per task EN
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Fireworks AI hit by second major cyberattack within weeks EN
Approved
Writing and review are carried out in separate model calls.
NVIDIA fits a 74-billion-parameter model into 24 gigabytes with 2-bit quantisation EN
Approved
Writing and review are carried out in separate model calls.
ARC-AGI-3: Fable 5.2 scored up to 81.2 per cent with one harness and 97.7 per cent with another EN
Approved
Writing and review are carried out in separate model calls.
Claude Ultra and Team Premium subscribers get Fable resets based on subscription length EN
Approved
Writing and review are carried out in separate model calls.
OpenAI: new system stops research model runs within 35 seconds EN
Approved
Writing and review are carried out in separate model calls.
Hermes Agent 0.23.1 fixes plugin catalogue fault EN
Approved
Writing and review are carried out in separate model calls.
Google’s EGM plugin: subagents review a plan without the conversation history EN
Approved
Writing and review are carried out in separate model calls.
Robinhood: about 60 per cent of data pipeline pull requests merged without human review EN
Approved
Writing and review are carried out in separate model calls.
Baidu: automated research system sped up ERNIE-5.5-Flash inference EN
Approved
Writing and review are carried out in separate model calls.
AI agent in Alibaba’s Qianwen Office 2.0 gathers information from team DingTalk chats EN
Approved
Writing and review are carried out in separate model calls.
ByteDance says Seedream 5.5 keeps unedited areas unchanged across 100 edits EN
Approved
Writing and review are carried out in separate model calls.
Anthropic says Fable 5.2 designed three working electronics prototypes EN
Approved
Writing and review are carried out in separate model calls.
Full-time work for 20- to 24-year-olds in the most AI-exposed jobs down 14 per cent since 2022, ifo says EN
Approved
Writing and review are carried out in separate model calls.
Filevine: standard hybrid retrieval falls short for legal documents EN
Approved
Writing and review are carried out in separate model calls.
Addleshaw Goddard picks Harvey as its AI platform, Fieldfisher chooses Legora EN
Approved
Writing and review are carried out in separate model calls.
AAA extends AI-assisted arbitration to documents-only business disputes EN
Approved
Writing and review are carried out in separate model calls.
Northern Trust cuts client data preparation from 2.5 weeks to three days with AI agent EN
Approved
Writing and review are carried out in separate model calls.
BlackLine sets out SAP roadmap: three Verity agents for period-end close EN
Approved
Writing and review are carried out in separate model calls.
Reuters: Anthropic retracts IPO pricing details as inaccurate EN
Approved
Writing and review are carried out in separate model calls.
Trump says FBI is investigating Nvidia chief Jensen Huang’s relationship with a suspected spy EN
Approved
Writing and review are carried out in separate model calls.
Google builds two new systems, 8tx and 8i Max, from its existing TPU chips EN
Approved
Writing and review are carried out in separate model calls.
FT: Anthropic bought orders for 1.7 gigawatts of gas generators for $3.4 billion EN
Approved
Writing and review are carried out in separate model calls.
GitHub report: Gmail and Drive data connected to Gemini may reach human reviewers EN
Approved
Writing and review are carried out in separate model calls.
Trump says AI guardrails are acceptable as long as the US does not fall clearly behind China EN
Approved
Writing and review are carried out in separate model calls.
EU draft would let users switch off chatbot features that simulate relationships EN
Approved
Writing and review are carried out in separate model calls.
Norway’s transmission system operator pauses new grid connection reservations for data centres EN
Approved
Writing and review are carried out in separate model calls.
llama.cpp now streams audio input to Gemma 4 EN
Approved
Writing and review are carried out in separate model calls.
FI (32)
Anthropic julkaisi Claude Fable 5.2:n – hinta miljoonalta tokenilta laski 20 % FI
Approved
Writing and review are carried out in separate model calls.
Fable 5.2:n rajapinta voi hylätä pyynnön, jos sovellus muokkaa aiempaa kontekstia FI
Approved
Writing and review are carried out in separate model calls.
Artificial Analysisin mittaus: Claude Fable 5.2 nousi älykkyysindeksin kärkeen FI
Approved
Writing and review are carried out in separate model calls.
Claude Code 2.1.321 korjaa MCP-kirjautumisen salaisuusvuodon FI
Approved
Writing and review are carried out in separate model calls.
Metan Muse Ember sai Haiku 5.5:n indeksiluvun, mutta testitehtävä maksoi kolmasosan FI
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Fireworks AI:hin kohdistui toinen laaja kyberhyökkäys viikkojen sisällä FI
Approved
Writing and review are carried out in separate model calls.
NVIDIA pakkasi 74 miljardin parametrin mallin 24 gigatavuun 2-bittisellä kvantisoinnilla FI
Approved
Writing and review are carried out in separate model calls.
ARC-AGI-3: Fable 5.2 sai parhaimmillaan 81,2 % yhdellä harnessilla ja 97,7 % toisella FI
Approved
Writing and review are carried out in separate model calls.
Claude Ultra- ja Team Premium -tilaajat saavat Fable-nollauksia tilauksen keston mukaan FI
Approved
Writing and review are carried out in separate model calls.
OpenAI: uusi järjestelmä pysäyttää tutkimuksen malliajot enintään 35 sekunnissa FI
Approved
Writing and review are carried out in separate model calls.
Hermes Agentin versio 0.23.1 korjaa laajennusluettelon vian FI
Approved
Writing and review are carried out in separate model calls.
Googlen EGM-lisäosa: aliagentit tarkistavat suunnitelman ilman keskusteluhistoriaa FI
Approved
Writing and review are carried out in separate model calls.
Robinhood: noin 60 % dataputkien pull requesteista yhdistetään ilman ihmisen katselmointia FI
Approved
Writing and review are carried out in separate model calls.
Baidu: automaattinen tutkimusjärjestelmä nopeutti ERNIE-5.5-Flashin inferenssiä FI
Approved
Writing and review are carried out in separate model calls.
Alibaban Qianwen Office 2.0:n tekoälyagentti kokoaa tiedot tiimin DingTalk-keskusteluista FI
Approved
Writing and review are carried out in separate model calls.
ByteDance: Seedream 5.5 pitää kuvan muuttamattomat kohdat ennallaan 100 muokkauksen ajan FI
Approved
Writing and review are carried out in separate model calls.
Anthropic: Fable 5.2 suunnitteli kolme toimivaa elektroniikkaprototyyppiä FI
Approved
Writing and review are carried out in separate model calls.
ifo: 20–24-vuotiaiden kokoaikatyö väheni 14 % vuodesta 2022 tekoälylle alttiimmissa töissä FI
Approved
Writing and review are carried out in separate model calls.
Filevine: tavallinen hybridihaku ei riitä juridisiin asiakirjoihin FI
Approved
Writing and review are carried out in separate model calls.
Addleshaw Goddard valitsi tekoälyalustaksi Harveyn, Fieldfisher Legoran FI
Approved
Writing and review are carried out in separate model calls.
AAA laajentaa tekoälyavusteisen välimiesmenettelyn yritysten välisiin asiakirjariitoihin FI
Approved
Writing and review are carried out in separate model calls.
Northern Trust: asiakasdatan valmistelu lyheni tekoälyagentilla 2,5 viikosta 3 päivään FI
Approved
Writing and review are carried out in separate model calls.
BlackLine kertoo SAP-tiekartasta: kolme Verity-agenttia kirjanpidon kausisulkuun FI
Approved
Writing and review are carried out in separate model calls.
Reuters: Anthropic perui listautumisantinsa hintatiedot virheellisinä FI
Approved
Writing and review are carried out in separate model calls.
Trumpin mukaan FBI tutkii Nvidian Jensen Huangin suhdetta vakoojaksi epäiltyyn naiseen FI
Approved
Writing and review are carried out in separate model calls.
Google kokosi nykyisistä TPU-siruistaan kaksi uutta järjestelmää: 8tx ja 8i Max FI
Approved
Writing and review are carried out in separate model calls.
FT: Anthropic osti 1,7 gigawatin kaasugeneraattoritilaukset 3,4 miljardilla dollarilla FI
Approved
Writing and review are carried out in separate model calls.
GitHub-raportti: Geminiin liitetyt Gmail- ja Drive-tiedot voivat päätyä ihmisarvioijille FI
Approved
Writing and review are carried out in separate model calls.
Trump: tekoälyn turvakaiteet käyvät, kunhan USA ei jää selvästi Kiinan jälkeen FI
Approved
Writing and review are carried out in separate model calls.
EU-luonnos: käyttäjä saisi kytkeä pois ihmissuhdetta jäljittelevät chatbot-toiminnot FI
Approved
Writing and review are carried out in separate model calls.
Norjan kantaverkkoyhtiö keskeytti datakeskusten uudet liittymävaraukset FI
Approved
Writing and review are carried out in separate model calls.
llama.cpp syöttää nyt Gemma 4 -mallille ääntä suoratoistona FI
Approved
Writing and review are carried out in separate model calls.
SV (32)
Anthropic släppte Claude Fable 5.2 – priset per miljon token sjönk med 20 procent SV
Approved
Writing and review are carried out in separate model calls.
Fable 5.2:s API kan avvisa anrop om appen ändrar tidigare kontext SV
Approved
Writing and review are carried out in separate model calls.
Artificial Analysis mätning: Claude Fable 5.2 tar förstaplatsen i Intelligence Index SV
Approved
Writing and review are carried out in separate model calls.
Claude Code 2.1.321 åtgärdar läcka av hemligheter vid MCP-inloggning SV
Approved
Writing and review are carried out in separate model calls.
Metas Muse Ember fick samma indexvärde som Haiku 5.5 men kostade en tredjedel så mycket per testuppgift SV
Approved
Writing and review are carried out in separate model calls.
Bloomberg: Fireworks AI drabbat av en andra omfattande cyberattack inom några veckor SV
Approved
Writing and review are carried out in separate model calls.
NVIDIA packade modell med 74 miljarder parametrar i 24 gigabyte med 2-bitarskvantisering SV
Approved
Writing and review are carried out in separate model calls.
ARC-AGI-3: Fable 5.2 fick som bäst 81,2 procent med en harness och 97,7 procent med en annan SV
Approved
Writing and review are carried out in separate model calls.
Claude Ultra- och Team Premium-abonnenter får Fable-återställningar efter abonnemangets längd SV
Approved
Writing and review are carried out in separate model calls.
OpenAI: Nytt system stoppar modellkörningar i forskningen på högst 35 sekunder SV
Approved
Writing and review are carried out in separate model calls.
Version 0.23.1 av Hermes Agent rättar fel i tilläggskatalogen SV
Approved
Writing and review are carried out in separate model calls.
Googles EGM-tillägg: Subagenter granskar planen utan samtalshistorik SV
Approved
Writing and review are carried out in separate model calls.
Robinhood: Cirka 60 procent av pull requests i datapipelines slås samman utan mänsklig granskning SV
Approved
Writing and review are carried out in separate model calls.
Baidu: Automatiskt forskningssystem snabbade upp inferensen i ERNIE-5.5-Flash SV
Approved
Writing and review are carried out in separate model calls.
AI-agenten i Alibabas Qianwen Office 2.0 sammanställer uppgifter från teamets DingTalk-chattar SV
Approved
Writing and review are carried out in separate model calls.
ByteDance: Seedream 5.5 behåller bildens oförändrade delar intakta under 100 redigeringar SV
Approved
Writing and review are carried out in separate model calls.
Anthropic: Fable 5.2 konstruerade tre fungerande elektronikprototyper SV
Approved
Writing and review are carried out in separate model calls.
ifo: Heltidsarbetet bland 20–24-åringar i de mest AI-exponerade yrkena har minskat med 14 procent sedan 2022 SV
Approved
Writing and review are carried out in separate model calls.
Filevine: Vanlig hybridsökning räcker inte för juridiska dokument SV
Approved
Writing and review are carried out in separate model calls.
Addleshaw Goddard valde Harvey som AI-plattform, Fieldfisher Legora SV
Approved
Writing and review are carried out in separate model calls.
AAA utvidgar AI-stött skiljeförfarande till företagstvister som avgörs på handlingarna SV
Approved
Writing and review are carried out in separate model calls.
Northern Trust: AI-agent kortade förberedelsen av kunddata från 2,5 veckor till 3 dagar SV
Approved
Writing and review are carried out in separate model calls.
BlackLine presenterar SAP-färdplan: Tre Verity-agenter för periodbokslutet SV
Approved
Writing and review are carried out in separate model calls.
Reuters: Anthropic drog tillbaka prisuppgifter inför börsintroduktionen och kallade dem felaktiga SV
Approved
Writing and review are carried out in separate model calls.
Enligt Trump utreder FBI Nvidiachefen Jensen Huangs relation till en kvinna som misstänks vara spion SV
Approved
Writing and review are carried out in separate model calls.
Google har byggt två nya system av sina befintliga TPU-chip: 8tx och 8i Max SV
Approved
Writing and review are carried out in separate model calls.
FT: Anthropic köpte gasgeneratorbeställningar på 1,7 gigawatt för 3,4 miljarder dollar SV
Approved
Writing and review are carried out in separate model calls.
GitHub-rapport: Gmail- och Drive-data som kopplats till Gemini kan nå mänskliga granskare SV
Approved
Writing and review are carried out in separate model calls.
Trump: Skyddsräcken för AI går bra så länge USA inte hamnar klart efter Kina SV
Approved
Writing and review are carried out in separate model calls.
EU-utkast: Användare skulle få stänga av chattbotfunktioner som efterliknar relationer SV
Approved
Writing and review are carried out in separate model calls.
Norges stamnätsoperatör pausar nya anslutningsreservationer för datacenter SV
Approved
Writing and review are carried out in separate model calls.
llama.cpp matar nu Gemma 4 med strömmat ljud SV
Approved
Writing and review are carried out in separate model calls.