The build, deploy and infrastructure-as-code tools teams live in.
Meta's new Muse Wonder model delivers comparable performance to Anthropic's Claude Fable 5.1 on agentic work, but consumed more tokens and ran up higher bills in head-to-head tests. Fable's safety filters also blocked routine backend code that Muse completed.
An Israeli startup emerging from stealth claims that AI agents repeatedly reconstructing company context from scattered systems waste enormous compute budgets. Uvi's shared memory platform consolidates that knowledge once, letting agents query it instantly across any frontier model.
Enterprises that prototype AI applications on developer-friendly databases often face expensive re-platforming when moving to production. pgEdge's leadership argues that selecting infrastructure capable of handling both stages from the start eliminates costly migration headaches.
Subquadratic has entered the decision-model market with SubQ Snap, claiming a 384K-token context window—12 times larger than Jev's 32K limit. The company argues that bigger context windows let decision models take on more substantial work, reducing reliance on expensive frontier LLMs.
SpaceXAI's new X Subscription consolidates Cursor, Grok, and X Premium into a single plan with five pricing tiers, but developers using competing models from Anthropic and OpenAI will face additional API charges starting from their first request.
Three months after entering public beta, OpenCode 2.0 is now the main release, featuring a shared background service that connects terminal, web and desktop interfaces alongside a rebuilt API and embeddable SDK.
GitHub Copilot works more effectively when it understands your project's conventions and workflows. Four customization approaches—repository instructions, agent skills, custom agents, and MCP server connections—let teams embed their standards directly into their codebase.
Cloudflare has made Transformers generally available, allowing teams to filter, redact, and restructure logs using SQL queries directly within the Logpush export pipeline instead of requiring separate processing infrastructure.
