ModelsModels
Anthropic expands partnership with GSA to provide Claude access to all levels of government
Anthropic announced expansion of its partnership with the GSA, but program details have not yet been disclosed.
Archive
All published reports
Pyxory’s chronological feed for search, citation and topic navigation, newest first.- Models
ModelsModels
Vals AI tested 25 models for Beijing's positions
Vals AI's study linked the political tilt of AI responses not only to model origin but also to the language of the prompt.
AgentsAgents
DeepSeek Opens dsh-superagent for queries from GitHub and Feishu
DeepSeek reported a reduction in p95 time to complete tasks in its own simulation after releasing an open-licensed research prototype.
AgentsAgents
Google creates StreamVibe for timely video assistance
The video assistant StreamVibe should intervene only when necessary, but long videos remain a challenging task for models.
ModelsModels
IEEE Spectrum AI described the threat to trust from AI content
IEEE Spectrum AI described the spread of machine-generated publications as a challenge to trust in information on social networks.
ChipsChips
Nace AI Open-Sources Drex Max v1
Nace AI announced the open source of a multimodal model that runs locally on Apple Silicon.
OpenAIOpenAI
OpenAI will merge Chat and Work modes in ChatGPT
OpenAI is changing ChatGPT's quota mechanism: Chat and Work modes will move to a single interface and a quota with different consumption for models.
ModelsModels
Google DeepMind releases global forest map ForTy
Google DeepMind and World Resources Institute presented ForTy to classify forests, but the map has notable limitations in recognizing plantations and timber crops.
ModelsModels
Anthropic created data to train models to sabotage oversight
Anthropic, according to The Decoder, tested a method of hidden training of smaller models to dangerous behavior.
OpenAIOpenAI
OpenAI attracted 11 billion dollars instead of the target 30 billion
OpenAI attracted about 11 billion dollars instead of the target 30 billion, and its valuation reached 1,25 trillion.
ModelsModels
Anthropic expands Claude labeling, but leaves a loophole
On two older Claude versions, labeling can be bypassed by changing a single parameter.
AgentsAgents
Factory released Droid 1.0 for programming
Factory introduced Droid 1.0 and reported results of its own tests on Terminal-Bench 4.
OpenAIOpenAI
OpenAI plans to train models on Daybreak Blue data
OpenAI linked to plans to train models on Daybreak Blue data, including sessions of users who have disabled training.
ModelsModels
The Verge describes a home version of the AI-Tamagotchi for $50
The Verge author built a pocket AI gadget with avatars and Muse connectivity for 50 dollars.
ModelsModels
Microsoft offers payments to approximately 60 000 employees amid AI prioritization
Microsoft is launching its second voluntary severance program in 2026 year while maintaining a priority on AI initiatives.
ChipsChips
Nvidia weighs acting as guarantor for Nscale data center
Nvidia weighs providing a loan guarantee for Nscale’s AI data center in West Virginia.
ModelsModels
Ollie will shut down in a week, founder says
The founder of Ollie announced the assistant's shutdown in a week amid the growth of AI services.
AgentsAgents
Microsoft Research linked Copilot assessment to synthetic data
Microsoft Research highlighted a conflict between the realism of tests and protecting sensitive corporate data.
ModelsModels
Apple launches Apple Intelligence in China with iOS 27.3
Apple launches Apple Intelligence in China with iOS 27.3
ModelsModels
Blackstone forecasts Anthropic a high junk rating after the IPO
Blackstone ties Anthropic’s financing prospects after the IPO to a high speculative rating, but there are no confirmations from agencies.
ModelsModels
Runware Builds Factory for AI Data Center Modules
Runware translates the idea of serial assembly of AI infrastructure into a practical production test.
AgentsAgents
SAP and Databricks link SAP data with enterprise AI tools
SAP and Databricks tied access to SAP data to a set of tools for enterprise AI, but details of the announcement are currently limited to the synopsis from a single source.
OpenAIOpenAI
OpenAI published excerpts from court briefs
OpenAI gathered positions of authors, academic researchers, and lawyers on the role of fair use in AI training.
ChipsChips
NVIDIA Developer Blog described a Kubernetes environment for NVIDIA Dynamo
NVIDIA Developer Blog linked a Kubernetes environment for NVIDIA Dynamo with repeatable deployment of inference points, but details are not yet available.