Social Media Briefing — April 24, 2026

Building on yesterday's Reddit buzz about an imminent OpenAI launch, OpenAI's primary GPT-5.5 announcement: 'A new class of intelligence for real work and powering agents.' Massive engagement (43.6K likes, 7.2M views).

Building on yesterday's Reddit buzz about an imminent OpenAI launch, Allie K Miller provides a detailed breakdown of OpenAI's GPT-5.5 launch, including benchmark results (82.7% Terminal-Bench 2.0, 84.9% GDPval, 85% ARC-AGI-2), enterprise use cases, availability tiers, and her assessment that we're entering a model class where improvements are less noticeable for non-technical users.

Building on yesterday's Reddit buzz about an imminent OpenAI launch, Ethan Mollick shares his detailed review of GPT-5.5 after weeks of testing, noting it conducted 'not-bad social science research on its own' and developed a novel RPG. He declares GPT-5.5 Pro as currently the best model for hard problems, while acknowledging continued 'jaggedness'. Links to his Substack essay.

Building on yesterday's Reddit buzz about an imminent OpenAI launch, Sam Altman's primary GPT-5.5 announcement tweet: 'GPT-5.5 is here!' — receives massive engagement (16.8K likes, 1.2M views).

Building on yesterday's Reddit buzz about an imminent OpenAI launch, Swyx provides a detailed breakdown of GPT-5.5 specs: 400K context in Codex, 1M in API, $5/$30 pricing, 20% self-improved inference speed, co-designed with GB200/GB300, and benchmark results across Terminal-Bench (82.7%), SWE-Bench Pro (58.6%), GDPval (84.9%), Tau2-bench (98.0%), BixBench (80.5%).

Building on yesterday's Reddit buzz about an imminent OpenAI launch, NVIDIA announces partnership with OpenAI on GB200 NVL72 systems for GPT-5.5, claiming 35x reduction in token costs. Notes GPT-5.5 is co-designed with GB200/GB300 and NVIDIA is using Codex Agents internally.

Building on yesterday's Reddit buzz about an imminent OpenAI launch, Simon Willison reports GPT-5.5 is out but NOT yet available in the OpenAI API. Notes that in the 'ongoing battle with Anthropic over OpenClaw,' OpenAI made their Codex tool's private backend API a semi-approved mechanism, and that one has GPT-5.5. Used it to draw pelicans as a test.

Ethan Mollick shares his review of GPT-5.5 after weeks of testing: it conducted passable social science research autonomously, developed a novel RPG, and GPT-5.5 Pro is currently the best model for hard problems despite continued 'jaggedness'.

xAI announces Grok Voice Think Fast 1.0, a voice model for complex multi-step workflows claiming #1 on Tau Voice Bench, with strong handling of noise, accents, and interruptions.

Following Social discussion from two days ago on open-source AI restrictions, Nathan Lambert provides a comprehensive analysis of proposed bans on Chinese AI distillation. Argues that while protecting US frontier labs is important, banning Chinese open-weight models would harm US startups (like Cursor), academic research, and the fragile open-source ecosystem. Warns of potential 6-12 month capability delay and consolidation of power to closed labs.

OpenAI officially announces GPT-5.5 rolling out for Plus, Pro, Business and Enterprise users in ChatGPT and Codex, plus GPT-5.5 Pro for Pro/Business/Enterprise.