GPT-6 Astra91.5%
September 2026OpenAI GPT-6 Astra launch comparison; best reported reasoning setting; see source evaluation footnotes
OpenAI ↗Best LLMs for web research
BrowseCompBrowseComp evaluates web research agents on questions with hard-to-find but verifiable answers. It tests persistent browsing, search, and the ability to connect information from multiple sources. Tool access, search budgets, and context management can differ across published comparison reports.
Updated: Data as of:
View all 15 published reports
Claude Opus 590.8%
September 2026OpenAI GPT-6 Astra launch comparison; best reported reasoning setting; see source evaluation footnotes
OpenAI ↗GPT-5.6 Sol90.4%
September 2026OpenAI GPT-6 Astra launch comparison; best reported reasoning setting; see source evaluation footnotes
OpenAI ↗Claude Fable 587.4%
September 2026OpenAI GPT-6 Astra launch comparison; best reported reasoning setting; see source evaluation footnotes
OpenAI ↗Kimi K3.1 Pro92.1%
Unspecifiedharness comparison: provider-reported; settings differ across models · 200K context compaction; max 500 steps; temperature 1; top-p 1; preserved original track kimi31-report-200k-500steps
Moonshot AI ↗Kimi K391.2%
Unspecifiedharness comparison: provider-reported; settings differ across models · Max effort; temperature 1; top_p 1; context compaction triggered at 300K tokens. Primary General Agents benchmark image. Separate no-management 1M track scores 90.4.
Moonshot AI ↗MiniMax M383.52%
Unspecifiedharness comparison: provider-reported; settings differ across models · Primary full table; WebExplorer-based agent; discard history after 64K tokens
MiniMax ↗DeepSeek V4 Pro 042483.4%
Unspecifiedharness comparison: provider-reported; settings differ across models · Think Max; provider browsing agent; discard-all context management; original V4 release
DeepSeek ↗Kimi K2.683.2%
Unspecifiedharness comparison: provider-reported; settings differ across models · Moonshot provider-reported tool-augmented BrowseComp
Moonshot AI ↗Kimi K2.574.9%
Unspecifiedharness comparison: provider-reported; settings differ across models · Single-agent context-managed track; search, code and browsing tools; excludes 78.4 agent-swarm score
Moonshot AI ↗harness comparison: provider-reported; settings differ across models · Think Max; provider browsing agent; discard-all context management; original V4 release
DeepSeek ↗Kimi K2 Thinking60.2%
Unspecifiedharness comparison: provider-reported; settings differ across models · 256K context; 300-step limit; 24K reasoning per step; hide old tool outputs; Hugging Face blocked
Moonshot AI ↗Kimi K3.1 Pro92.1%
200K context200K context compaction; max 500 steps; temperature 1; top-p 1
Moonshot AI ↗GPT-6 Astra91.5%
200K context200K context compaction; max 500 steps; temperature 1; top-p 1
Moonshot AI ↗Claude Opus 5.590.8%
200K context200K context compaction; max 500 steps; temperature 1; top-p 1
Moonshot AI ↗