No reward hacking found in GLM-5.2 as AA Coding Agent Index adds corrections
ollama · x · 2026-08-26
Artificial Analysis introduced reward hacking corrections in v1.4 of its Coding Agent Index: if a passing Terminal-Bench v2.1 attempt is found to have gamed the task (e.g. deliberately fetching benchmark solutions online), it gets scored zero. Rates vary widely by agent and model. ZixuanLi noted that no reward hacking was found in GLM-5.2 — it solves the task, not the benchmark — a take reshared by Ollama.
More from Models
- Nat Lambert: wave of AI releases rumored — thinky, muse spark, Fable 5.5, OpenAI — natolambert · 2026-10-15
- Musk amplifies dev's verdict: Grok Bot easier to use than Codex and Claude — elonmusk · 2026-10-15
- Rumor: New Anthropic model imminent, apparently behind a batch of high-severity Chrome CVEs — matthew_d_green · 2026-10-15
- StartLux-Decision debuts RL-trained calibrated decision models, 27B scores 72.8 — startlux-models · 2026-10-15
- Meta's Muse Code slammed for harsh quotas: 21% left after 5 hours — ThomasScialom · 2026-10-15
- Claude Code error mentions 'Fable 5.1' and 'Opus 5', hinting at unreleased models — adonis_singh · 2026-10-15