Atlas

Benchmarks

← All benchmarks

Public Benefits Bench v1 - Multi Turn

Professional Work · 2026-06-09

The Multi Turn split of Public Benefits Bench v1 allows conversation across turns without web search. It is kept under the archived v1 benchmark because v1.1 publishes a different leaderboard without these four condition splits.

Top models (higher is better)

ModelScore
Claude Fable 565.2
Opus 4.858.5
Gemini 3.5 Flash54.0
GPT-5.553.9
Gemini 3.1 Pro Preview52.9
Sonnet 4.652.4
DeepSeek-V4-Pro48.7
MiniMax M348.2
GLM-5.147.6
Grok 4.346.9
Kimi K2.646.3
Grok 4.1 Fast41.7
Haiku 4.533.9
Loading Atlas data…