Agent Arena: OpenAI and Anthropic Lead on User Enjoyment as Labs Converge on Reliability
arena · x · 2026-10-17
Agent Arena tracked each lab's best agent (Anthropic, OpenAI, Google DeepMind, Moonshot, xAI, DeepSeek) across 4 signals from late June to early October. Praise vs Complaint shows OpenAI and Anthropic at +35 over the leaderboard average vs +513 for Moonshot/xAI/DeepSeek — the widest split. Confirmed Success jumped from -4 to +14 in three months, led by Claude Opus 5.5 (Max) at +20. Error recovery is converging: Bash Recovery spread halved (34→16), Steerability nearly halved (19→11). Labs are catching up on avoiding failures; user enjoyment is where the top three pull away.
More from coding & agent
- StackOne ships Tool Policies: fine-grained guardrails for MCP tool calls — afp-media · 2026-10-17
- Huntley: developers displaced by AI 'failed a curiosity test' — rseroter · 2026-10-17
- Teaching may be the best prerequisite for prompting agents, argues Geoffrey Litt — sh_reya · 2026-10-17
- RAX open-sources everything: zero demos, all experiment sessions included — du_yilun · 2026-10-17
- RAX separates the coding agent workspace from the robot stack for multi-embodiment — du_yilun · 2026-10-17
- shadcn reflects: it's only been a year since Claude Skills launched — shadcn · 2026-10-17