BrowseComp leaderboard
As of November 8, 2026, Kimi K3.1 Pro (Moonshot) leads BrowseComp at 92.1%, ahead of GPT-6 Astra at 91.5% and Kimi K3 at 91.2%. 6 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.
What is BrowseComp? BrowseComp tests how well a browsing agent can find a hard-to-locate fact on the web: 1,266 questions whose short answers are buried in tangled information but easy to check. Made by OpenAI, April 2025.
Top models on BrowseComp, November 2026
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | Kimi K3.1 Pro Moonshot (Kimi) | max effort | $2 / $10 | 1M | Vendor-reported Moonshot AI |
| 2 | GPT-6 Astra OpenAI | as reported | $10 / $50 | 1.05M | Vendor-reported OpenAI |
| 3 | Kimi K3 Moonshot (Kimi) | max effort | $3 / $15 | 1M | Vendor-reported Moonshot AI |
| 4 | GPT-5.6 Sol OpenAI | as reported | $4 / $20 | 1.05M | Vendor-reported OpenAI |
| 5 | Gemini 3.1 Pro Preview Preview Google · listed as “Gemini 3.1 Pro” | search, Python and browse, thinking high | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 6 | Gemini 3 Pro Preview Retired Google · listed as “Gemini 3 Pro” | search, Python and browse, thinking high | $2 / $12 | 1M | Vendor-reported Google DeepMind |
BrowseComp: score against price
Can you trust BrowseComp scores?
No independent leaderboard publishes BrowseComp results for these models yet, so every number on this page is the lab's own report of its own model. Labs choose the harness, the effort level and the subset they report, and they rarely publish failures. Read the setting under each score before comparing two rows.
- Questions were filtered to be unsolvable by OpenAI's own models at the time they were written.
- Extra attempts and test-time compute raise the score, so a number depends on the sampling budget.
- There is no official leaderboard; labs report their own runs with their own tools.
How BrowseComp works
The task
A fact-seeking question with a single short answer; the agent has to search and browse persistently to find it.
The score
Accuracy, with a model grader checking whether the answer means the same as the reference.
Test set
1,266 questions, as published by the authors.
Made by
OpenAI, first published April 2025.
Official sources
- openai.com/index/browsecomp
- Paper: arxiv.org/abs/2504.12516
- Definition read on 2026-11-08
Sources and freshness
6 published scores from 4 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.
- Vendor-reported Google DeepMind
- Vendor-reported OpenAI
- Vendor-reported Moonshot AI
- Vendor-reported Moonshot AI
BrowseComp FAQ
Answers derived from the 6 published scores on this page and the current price index.