← All benchmarks
FrontierChallenge (Pass Rate, Hard tasks)
Science · 2026-08-25
The share of FrontierChallenge's 74 Hard scientific-workflow tasks that a configuration fully completes, in percent, from the same single run per task as the 97-task Pass Rate and under the same historical rule (a native score of at least 99.9 out of 100, not recomputed under the September 23, 2026 policy of a completed task_score above 0.999). With the 23 Medium tasks it makes up the overall Pass Rate.
Benchmark source: arXiv
Top models (higher is better)
| Model | Score | Source |
|---|
| GPT-5.6 Sol | 14.9 | FrontierChallenge paper Table 2 (GPT-5.6 Sol / Codex, Hard) · Apodex |
| Grok 4.6 | 13.5 | FrontierChallenge paper Table 2 (Grok 4.6 / Claude Code, Hard) · Apodex |
| Claude Opus 5 | 10.8 | FrontierChallenge paper Table 2 (Claude Opus 5 / Claude Code, Hard) · Apodex |
| Kimi K3 | 10.8 | FrontierChallenge paper Table 2 (Kimi K3 / Claude Code, Hard) · Apodex |
| Qwen3.8-Max | 9.5 | FrontierChallenge paper Table 2 (Qwen 3.8 Max / Claude Code, Hard) · Apodex |
| DeepSeek V4 Pro 0813 | 8.1 | FrontierChallenge paper Table 2 (DeepSeek V4 Pro-0813 / Claude Code, Hard) · Apodex |
| GPT-5.6 Terra | 8.1 | FrontierChallenge paper Table 2 (GPT-5.6 Terra (max) / Codex, Hard) · Apodex |
| Apodex 1.1 | 6.8 | FrontierChallenge paper Table 2 (Apodex 1.1 / Frontier Agent (Agent Team), Hard) · Apodex |
| Gemini 3.7 Flash | 6.8 | FrontierChallenge paper Table 2 (Gemini 3.7 Flash / Claude Code, Hard) · Apodex |
| DeepSeek-V4-Flash-0731 | 5.4 | FrontierChallenge paper Table 2 (DeepSeek V4 Flash-0731 / Claude Code, Hard) · Apodex |
| GLM-5.2 | 2.7 | FrontierChallenge paper Table 2 (GLM-5.2 / Claude Code, Hard) · Apodex |
| Qwen3.5 397B A17B | 1.4 | FrontierChallenge paper Table 2 (Qwen3.5-397B-A17B / Claude Code, Hard) · Apodex |