Atlas

Benchmarks

← All benchmarks

FrontierChallenge (Avg. Score)

Science · 2026-08-25

FrontierChallenge's Avg. Score: the mean of the task-specific graders' 0-100 scores over all 97 released scientific-workflow tasks, partial credit included, from the same single run per task as the Pass Rate. The paper treats it as a descriptive measure of partial progress rather than a calibrated cross-task scale, because the rubrics differ by task; it prints the values in a chart and, for some configurations, in its text, and the full values are in the leaderboard page's data.

Benchmark source: arXiv

Top models (higher is better)

ModelScoreSource
GPT-5.6 Sol87.9FrontierChallenge leaderboard data, script.js results (GPT-5.6 Sol / Codex, avgScore) · Apodex
Grok 4.686.6FrontierChallenge leaderboard data, script.js results (Grok 4.6 / Claude Code, avgScore) · Apodex
Kimi K385.5FrontierChallenge leaderboard data, script.js results (Kimi K3 / Claude Code, avgScore) · Apodex
Claude Opus 584.9FrontierChallenge leaderboard data, script.js results (Claude Opus 5 / Claude Code, avgScore) · Apodex
GPT-5.6 Terra84.7FrontierChallenge leaderboard data, script.js results (GPT-5.6 Terra (max) / Codex, avgScore) · Apodex
Qwen3.8-Max82.5FrontierChallenge leaderboard data, script.js results (Qwen 3.8 Max / Claude Code, avgScore) · Apodex
DeepSeek-V4-Flash-073181.5FrontierChallenge leaderboard data, script.js results (DeepSeek V4 Flash-0731 / Claude Code, avgScore) · Apodex
DeepSeek V4 Pro 081381.4FrontierChallenge leaderboard data, script.js results (DeepSeek V4 Pro-0813 / Claude Code, avgScore) · Apodex
Apodex 1.174.5FrontierChallenge leaderboard data, script.js results (Apodex 1.1 / Frontier Agent (Agent Team), avgScore) · Apodex
Gemini 3.7 Flash73.4FrontierChallenge leaderboard data, script.js results (Gemini 3.7 Flash / Claude Code, avgScore) · Apodex
Qwen3.5 397B A17B67.8FrontierChallenge leaderboard data, script.js results (Qwen3.5-397B-A17B / Claude Code, avgScore) · Apodex
GLM-5.267.5FrontierChallenge leaderboard data, script.js results (GLM-5.2 / Claude Code, avgScore) · Apodex
Loading Atlas data…