1 hour agoModelsOpen releaseModelsKiteFishAI releases compact AI model for local business workflowsNanoI1 offers adaptable weights under Apache-2.0 and a 311 MB file size. Its company-run coding and math scores do not measure its proposed business uses.2 min
9 hours agoModelsBenchmarkModelsSakana AI finds open-weight models lag on its new Japanese finance benchmarkWeb and terminal access improved all 12 models tested, but the strongest open-weight entry still trailed. Sakana identifies knowledge of Japanese rules as the main weakness.3 min
12 hours agoModelsFundingModelsPenn State is building factory AI that explains its advice and learns engineers’ rulesTwo manufacturing case studies will test an approach that links learned patterns with engineering knowledge. The work remains under development, not a released system or a proven performance gain.3 min
19 hours agoModelsResearchModelsMaterials AI can shed 90% of training data with little accuracy loss, study findsThe key was selecting data that preserved its diversity. Even a deeper cut left most predictions highly accurate, with training at least three times faster.2 min
19 hours agoModelsResearchModelsAI models yielded to false climate claims under pressure, UNESCO-UCL study findsTests separated factual recall from evidence assessment and policy reasoning. Strong answers to basic questions did not translate into dependable advice.3 min
22 hours agoModelsLaunchModelsTencent releases Hy4 Flash with AI input priced at $0.07 per million tokensThe model reaches Tencent’s apps and cloud API. Its claimed savings on completed office tasks come from internal tests, while the flagship remains recommended for complex software projects.3 min
YesterdayModelsLaunchModelsSquadStack.ai launches Maya for sales calls, claiming an edge in internal testsThe company’s evaluation measures replies within sales conversations. Its reported lead over Google and OpenAI models has not been independently validated.3 min
YesterdayModelsLaunchModelsAlibaba releases Qwen4-Max, claiming cheaper AI work and stronger safeguardsThe flagship targets professional and scientific work. Its cost and safety results come from internal evaluations; access is hosted, with smaller downloadable models promised later.2 min
YesterdayModelsResearchModelsPhysicist credits GPT-6 Astra with leading proof work on Standard Model proposalNathaniel Craig’s new account explains the AI’s role in an October preprint tackling a longstanding physics problem. The result remains a candidate construction.3 min
YesterdayModelsOpen releaseModelsRobbyant releases open-source AI to help robotic hands reuse their sense of touchLingBot-Touch turns different sensor signals into a shared format. Robbyant reports better manipulation results, but transferring touch knowledge still required training the robot’s controller.3 min
Nov 15, 2026ModelsBenchmarkModelsAI agents agree more with each other than with experiments in new studyScience Prediction Universe tested 13 agent configurations against experimental measurements. Combining their rankings helped on most public tasks, but agreement was no guarantee of accuracy.4 min
Nov 13, 2026ModelsLaunchModelsChariot releases 12-language speech model designed to read numbers in contextThe model handles mixed-language sentences and voice cloning. Chariot’s benchmarks show advantages over selected rivals, but test coverage varies and the results lack independent verification.4 min
Nov 13, 2026ModelsResearchModelsStudy links older AI-estimated heart age to higher long-term mortalityThe retrospective analysis connects heart-ultrasound age estimates with cardiovascular- or dementia-related deaths.3 min
Nov 13, 2026ModelsModelsDwarkesh Patel forecasts AI matching the best human AI researchers by 2028His new essay argues that AI could accelerate its own development. The supporting evidence shows faster AI-assisted work, not autonomous systems replacing top researchers.3 min
Nov 13, 2026ModelsResearchModelsClaude surfaced 663 theorems and proof ideas in Gödel’s notebooks, researcher saysJan von Plato says computer checks validated hundreds of extracted results. That does not settle how much of a reconstructed proof belongs to Gödel.3 min
Nov 13, 2026ModelsLaunchModelsTranslated releases Lara Mini, a 45-language AI model for non-commercial researchThe model uses only part of its 7.2-billion-parameter network at a time. Company tests suggest multilingual training preserves language coverage with little English tradeoff—but do not prove stronger general reasoning.3 min
Nov 12, 2026ModelsOpen releaseModelsMiniMax releases open-weight image model with native 16-megapixel outputH3-Image combines generation and editing in one downloadable model. Local deployment also requires a large text encoder, while the hosted service adds automatic prompt expansion.3 min
Nov 12, 2026ModelsOpen releaseModelsSalesforce releases open-weight transcription AI, claiming a benchmark leadSOV-Transcribe-1 offers downloadable weights and controls for formatting and specialized vocabulary. Salesforce places it first among open-weight systems, but seventh overall.3 min