What mattered last week in AI

Anthropic drops Sonnet 5

Anthropic released Sonnet 5 alongside the return of Fable 5 — the company's first major model update since the Claude 3.5 family, with improvements in coding, reasoning, and multimodal capabilities. Early benchmarks show gains on agentic tasks. The model arms race isn't slowing. If you're coding, Sonnet 5 is now a better option to benchmark against.

Seltz builds search for agents

Seltz, a startup rebuilding web search for AI agents, raised $12.5 million in seed funding. The company is building a search API specifically designed for agentic workflows — not for humans, but for bots that need to navigate and retrieve from the web. The infrastructure for agents is being built right now. If you're shipping agents, Seltz is one to watch.

OpenAI releases GPT-5.4-Cyber

OpenAI released GPT-5.4-Cyber in a limited rollout — following Anthropic's lead on constrained model releases. The cyber-focused variant targets security and coding tasks. Access remains restricted to select partners. The model arms race is being replaced by the release-strategy arms race. OpenAI also acquired Ona, a cloud startup, to boost its Codex AI coding tool. The acquisition adds engineering talent and infrastructure directly into OpenAI's developer-tools stack. Terms were not disclosed. OpenAI wants to own the coding workflow end-to-end. Limited drops create scarcity and startup acquisitions create momentum.

Meta promises coding model

Meta announced it will release a new version of its flagship Muse Spark AI model with advanced coding capabilities 'soon.' The model is expected to close the gap with OpenAI and Anthropic on code generation and software engineering tasks. Muse Spark includes a "contemplating mode" that uses AI agents to improve the quality of prompt responses. No specific date or name was given. Another major AI player entering back in the coding-model race. The benchmark keeps rising — your agents get smarter without you lifting a finger.

Assort Health raises $120M & Venice AI raises $65M

Assort Health closed a $120 million Series C to scale its voice AI agent platform for healthcare. The platform handles patient interactions, scheduling, and clinical triage through voice-based agents. This is one of the largest fundraises this year for vertical AI agents. Vertical AI agents are attracting real money. Voice + healthcare is proving there's product-market fit outside the general-purpose stack. Venice AI raised $65 million in Series A funding to build privacy-focused AI infrastructure. The startup positions itself as an alternative to Big Tech AI clouds with local-first, encrypted inference. Privacy-as-differentiator is no longer niche. It's a requirement for enterprises that can't afford data leakage.

Trump restrictions boost open source

Trump's administration restricted access to private frontier AI models, pushing developers toward open source alternatives. The policy shift signals that closed models face regulatory headwinds while open weights gain strategic value. The policy wind is at open source's back. Build on platforms you control, not ones that can be restricted.

Qualcomm buys Modular for $3.9B

Qualcomm acquired AI software startup Modular for $3.9 billion — one of the largest AI software acquisitions this year. Modular's platform lets developers build and deploy AI models across devices. The deal gives Qualcomm a full-stack AI software layer to pair with its chips. The hardware play is done. Now the race is for the software stack that runs on every device.

From Sugata AI

SUGATA AI releases HomeSky

Running a Bluesky account by hand doesn't scale. So the team at SUGATA AI built HomeSky, an automation platform for Bluesky. Schedule posts, map your Bluesky network, growth center, clean-up, auto-post from RSS feeds and much more — all from one AI-powered dashboard. No manual busywork. Free to start, if you're on Bluesky.

Building with AI?

We turn this stuff into shipped software — agents, MCP tools, automation, and apps.