Anthropic recently hosted its inaugural "Code with Claude" conference, showcasing its managed coding harness at full scale. Key announcements included doubling Claude Code's five-hour rate limits for Pro, Max, Team, and seat-based Enterprise plans, removing peak-hour reductions, and raising Opus API limits. Additionally, Anthropic secured a deal with SpaceX for the full capacity of the Colossus 1 data center, encompassing over 300 megawatts and 220,000 Nvidia GPUs scheduled to come online within the month. Product updates introduced multi-agent orchestration and Outcomes to public beta for Managed Agents, while "dreaming" for self-improving memory landed in research preview. Boris Cherny also demonstrated remote agents and routines, transforming Claude Code into an asynchronous workflow engine.
This official progress, however, represents only half the story. The other half involves OpenCode, an open-source coding agent that Anthropic has spent the last three months attempting to disconnect from Claude. OpenCode has now become the most-starred coding harness on GitHub, boasting 157,000 stars on its SST OpenCode repository, significantly surpassing Anthropic's official anthropics/claude-code which holds approximately 122,000 stars as of this week.
These two distinct narratives are not unfolding simultaneously but rather on a convergent trajectory, a trend warranting closer examination.
The genesis of this situation dates back to January, not May. On January 9, 2026, Anthropic deployed server-side checks that effectively blocked third-party tools from authenticating to Claude Pro and Max subscriptions via OAuth. Tools impacted included OpenCode, Cline, and RooCode. These tools had been sending HTTP headers that led Anthropic's servers to believe requests originated from the official Claude Code binary, allowing users to run autonomous agent workflows on a flat $200 monthly Max subscription instead of incurring substantially higher API rates for equivalent token consumption.
Anthropic's rationale for the block is defensible: subscriptions are intended to subsidize first-party usage. Routing the same workloads through subscription OAuth tokens via a third-party harness transforms this subsidy into a cost center, offering no benefit to the funding company. The Hacker News discussion on the OpenCode submission features numerous developer comments acknowledging this perspective, including those who later starred the repository. Anthropic itself clarified this policy in March, stating that subscription OAuth tokens were never intended for routing requests via third-party products.
Developers' primary grievance, however, centered on the execution: a complete lack of advance notice. Some accounts were banned due to triggering abuse filters during the cutover. The block was implemented at 02:20 UTC, interrupting workflows for users in various time zones.
In response, the OpenCode team swiftly implemented ChatGPT Plus support within hours and began broadening its provider coverage. Anthropic, conversely, formalized its policy in updated Terms of Service (TOS) by February 19.
