SkillEvolBench (SelfGen-Revision) - Deployment Success: leaderboard
Skill reuse by coding agents on SkillEvolBench (Lei et al., 2026): 180 verifier-scored tasks in six environments such as code repair, API orchestration and document transformation, each model running in its vendor agent harness (Claude Code, Codex CLI or Gemini CLI). The agent starts with no skills, induces them from its own acquisition trajectories and verifier feedback and revises them after failures. After the acquisition tasks the library is frozen, and the board scores the 90 deployment tasks, which embed a procedure in a broader request, add shortcut traps or require composing skills; success rate in percent, higher is better.
Metric: Frozen deployment success rate (%; ESR, share of the 90 context-shift, adversarial and composition tasks, 30 of each, solved with the environment's frozen skill library; skills the agent induces from its own acquisition trajectories and verifier feedback, revised after failures; verifier-scored). Source: arxiv.org. Saturation forecast: Not forecast. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.6 | 40 | #158 |
| 2 | Claude Sonnet 4.5 | 35.6 | #244 |
| 3 | GPT-5.2 Codex | 34.4 | #185 |
| 4 | Gemini 3 Flash | 33.3 | #159 |
| 5 | Gemini 3.1 Pro (Preview) | 32.2 | #91 |
| 6 | Claude Opus 4.6 | 32.2 | #100 |
| 7 | GPT-5.4 | 31.1 | #121 |
| 8 | Claude Opus 4.5 | 31.1 | #152 |
| 9 | GPT-5.3 Codex | 31.1 | #114 |
| 10 | Gemini 2.5 Pro | 20 | #262 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=skillevolbench-selfgen-revision-deployment-success · How It Works · Data refreshed daily, snapshot 2026-11-12.