Social Media Briefing — April 17, 2026

Greg Brockman announces GPT-Rosalind, OpenAI's frontier model specifically designed for life science research, aimed at accelerating science and improving human outcomes.

Building on Reddit buzz from earlier this week, Anthropic announces that Opus 4.7 uses more thinking tokens, so rate limits have been increased for all subscribers to compensate. Massive engagement (829K views, 16.6K likes).

OpenAI's main announcement tweet for GPT-Rosalind: a frontier reasoning model for biology, drug discovery, and translational medicine.

Perplexity officially launches Personal Computer - an AI agent that integrates with the Mac app for orchestration across local files, native apps, and browser. Rolling out to Max subscribers and waitlist.

Building on Reddit buzz from earlier this week, Boris Cherny (Anthropic/Claude Code) launches a major thread sharing tips for getting the most out of Claude Opus 4.7, which he's been dogfooding for weeks. The thread received massive engagement (~972K views).

OpenAI's main Codex announcement tweet: it can now use Mac apps, connect to more tools, create images, learn from actions, remember preferences, and handle ongoing/repeatable tasks.

bcherny details Claude Code's new auto mode for Opus 4.7 - permission prompts are routed to a model-based classifier instead of requiring manual approval, enabling parallel Claude sessions. Available for Max/Teams/Enterprise.

OpenAI announces GPT-Rosalind Life Sciences model series is available as research preview for qualified customers including Amgen, Moderna, Allen Institute, and Thermo Fisher Scientific.

bcherny describes Opus 4.7 as more intelligent, agentic, and precise than 4.6, noting it took a few days to learn how to work with it effectively. Links to blog post.

Building on Reddit buzz from earlier this week, Announcing Opus 4.7 in Claude Code: more agentic, more precise, better at long-running work, carries context across sessions, handles ambiguity better.

Building on yesterday's Reddit sighting on Google Vertex, Scobleizer publishes a comprehensive AI-agent-generated report comparing Claude Opus 4.7 vs GPT-5.4 vs Gemini 3.1 Pro, covering hallucinations, coding, long context, and writing quality. Key findings: Opus 4.6 hallucination degradation was linked to Anthropic lowering default reasoning compute; Opus 4.7 has confirmed long-context regression (MRCR dropped from 78.3% to 32.2%); Gemini 3.1 Pro leads on hallucination benchmarks by 22 points over Claude; GPT-5.4 preferred for deep coding; Claude Mythos expected to leapfrog all.

Sam Altman announces major Codex improvements, highlighting computer use as a key update - it can use all Mac apps in parallel without interfering with direct user work.