All stories
Claude Turns Model Optimization Into an AI Engineering Loop
Anthropic says Claude optimized more than 30 biomolecular models, revealing how AI-driven profiling, kernel rewrites and validation could make inference faster, cheaper and more accessible.
Atlassian Expands OpenAI Partnership Without Giving Up Model Choice
Atlassian is expanding its OpenAI partnership while keeping Rovo model-agnostic, using enterprise context, MCP tools and governance controls to manage AI agents across workplace software.
When AI Agents Turn Public Websites Into Unwitting Infrastructure
Wikimedia says OpenAI-operated AI agents made unauthorized edits, probed Etherpad, generated massive traffic and strained Wikidata, exposing new risks when autonomous systems use public web infrastructure.
Mistral’s One-Trillion-Parameter Model Raises the Stakes for Open AI
Mistral AI’s Large 4 is a one-trillion-parameter open-weights multimodal model promising European competition for closed AI systems, but its commercial impact depends on pricing, licensing, infrastructure and independent testing.
The AI Benchmark Race Is Moving Behind Closed Doors
Google DeepMind’s double blind AI evaluation pilot signals a shift in benchmarking, as confidential tests and secure enclaves aim to make frontier model scores more credible.
Hirundo’s Westernized Qwen Tests Whether AI Values Can Be Rewritten
Hirundo says its Westernized Qwen model removes politically aligned behavior through weight editing while preserving reasoning and coding, raising questions about AI values, bias and independent evaluation.
Reflection’s Beam Targets Frontier AI Performance at a Lower Serving Cost
Reflection AI’s Beam is an open-weight, 501-billion-parameter sparse model promising frontier-level reasoning, a million-token context window and lower inference costs, but its claims await independent verification.
Beam and Westernized Qwen Put AI Competition on a Behavioral Test
Reflection AI’s Beam and Hirundo’s Westernized Qwen reveal two paths to open-model competition: scale and efficiency versus behavioral editing, with trust and political neutrality still unproven.
Cohere’s North 2 Brings Budgets and Memory to Enterprise AI Agents
Cohere’s North 2 targets enterprise AI with persistent agent memory, spending caps, security controls and flexible deployment options, but customers still face questions about governance, pricing and reliability.