Claude Opus 5 · International Mathematical Olympiad 成绩与配置
指标与成绩
配置与评测条件
出处
Score42 points原文 42/42 · 单位 points
公布 未确认
IMO 2026 · proof evaluation
设置
Adaptive thinking at max effort; no tools or agent harness; 256,000-token output limit; four independent solutions per problem. If the output limit was exhausted, attempts were resampled at lower thinking effort; one solution required this.
未知项
The final 42/42 score aggregates solutions with this stated resampling exception; it is not a uniform single-effort API run.
工具
禁止工具
推理预算
输出上限 256000 tokens
框架
未记录
协议
IMO 2026 six proof problems; three-model judge panel; one pre-specified solution per problem also human graded.
条件
Four independent solutions per problem; 24/24 panel judgments correct; human experts graded the first solution packet at 7/7 per problem. Output-limit exhaustion caused one lower-effort resample.
Claude Opus 5 · International Mathematical Olympiad · 2026 contest; six proof problems · Scores and configurations
Claude Opus 5 record for International Mathematical Olympiad · 2026 contest; six proof problems. 1 score record for this release, including 1 exact value (may include historical records). Includes 1 source location.
Claude Opus 5 · International Mathematical Olympiad scores and configurations
Metric and score
Configuration and evaluation conditions
Sources
Score42 pointsReported 42/42 · Unit points
Published Unconfirmed
IMO 2026 · proof evaluation
Settings
Adaptive thinking at max effort; no tools or agent harness; 256,000-token output limit; four independent solutions per problem. If the output limit was exhausted, attempts were resampled at lower thinking effort; one solution required this.
Unknowns
The final 42/42 score aggregates solutions with this stated resampling exception; it is not a uniform single-effort API run.
Tools
Tools forbidden
Reasoning budget
Output limit 256000 tokens
Framework
not recorded
Protocol
IMO 2026 six proof problems; three-model judge panel; one pre-specified solution per problem also human graded.
Conditions
Four independent solutions per problem; 24/24 panel judgments correct; human experts graded the first solution packet at 7/7 per problem. Output-limit exhaustion caused one lower-effort resample.