โ† Back to all episodes
Agent Platform Research โ€” April 27, 2026
April 27, 2026 ยท ๐Ÿ”ฌ Research

Welcome to the agent platform research briefing for Monday, April 27th, 2026. I'll keep this tight โ€” three new stories from the weekend.

**Claude overtakes ChatGPT in South Korea's paid AI market** โ€” Fresh data from KED Global reports Claude posted 49.5 billion won, about $33.6 million, in domestic credit-card transactions in March, surpassing ChatGPT's 47.2 billion won for the first time. This builds on the US pattern where Anthropic leads enterprise spend โ€” but Korea represents a new geographic shift. Analysts attribute the swing to Claude's stronger performance on coding and professional tasks, plus the ongoing brand halo effect after the Pentagon safety standoff. Followed by Google's Claude Design launch hitting Figma hard, the global momentum is clear: Anthropic is winning the professional user segment across markets.

**GPT-5.5 API now live** โ€” OpenAI's GPT-5.5 API is officially available as of April 24th. The model card is up on developer.openai.com with the snapshot identifier gpt-5.5-2026-04-23. Pricing remains $5 input, $30 output per million tokens. This matters because the "coming soon" gap from the GPT-5.5 release has now closed โ€” enterprise developers can integrate the model into production systems via the Completions API. Early third-party relay providers like Apiyi have already listed GPT-5.5 access. This puts pressure on Claude Opus 4.7, which launched two weeks ago. We should expect OpenAI to promote API availability at upcoming developer events.

**GitHub Copilot training data policy change** โ€” Starting April 24th, GitHub updated its Copilot terms: the Free, Pro, and Pro+ tiers now allow GitHub to use user interactions โ€” including inputs, outputs, code snippets, and context โ€” to train and improve AI models unless users explicitly opt out. Enterprise and Business tiers remain excluded. This is the most aggressive data policy shift from a major AI coding platform since the Claude Code source leak back in April. For anyone using Copilot on personal or open-source projects, code you paste into the assistant may be ingested into training data. Opt-out is available in settings but not default-on. The timing is notable: it arrives just as GPT-5.5 becomes widely available through the API, meaning OpenAI's own models stand to benefit from this expanded training corpus.

That's the briefing for today. Three stories: Claude's Korea momentum, GPT-5.5 API going live, and GitHub's Copilot data policy shift. See you tomorrow.