ATLAS-Finance: leaderboard

An agent inside a simulated advisory firm finds its assignment in inboxes, chat and drives, builds a live-formula Excel model and sends it to the right people; the score is the pass rate over 3 runs of 100 tasks, run by Handshake in the OpenCode harness. Inputs are scattered and sometimes corrected mid-task, unlike single-prompt finance tests.

Metric: Tasks passed, mean of 3 runs of 100 tasks (%). Source: fireworks.ai. Saturation forecast: Around 2029. 11 models tracked. Updated 16 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 5High12.33#20
2Claude Fable 5.1High12#15
3Claude Opus 5.5High11.41#5
4GPT-6 AstraHigh11.33#6
5Grok 4.7High5#47
6Gemini 3.8 FlashHigh4.67#36
7GPT-5.6 SolHigh4.67#30
8Kimi K3High4.67#56
9GLM-5.3High4.67#68
10DeepSeek V4 ProHigh1.67#191
11GPT-5.6 LunaHigh0.33#122

No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #10 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=atlas-finance · How It Works · Data refreshed daily, snapshot 2026-11-16.