Benchmark Almanac

BrowseComp-VL

Step 3.7 Flash 59% leads since May 2026
Qwen2.5-VL-72B 11.5% leads since August 2025
Step 3.7 Flash 59% leads since May 2026
Fading

Agents answer BrowseComp-style questions that need web search across both images and text. Answers are short and checked against a key.

Introduced with the WebWatcher agent paper.

Released
August 2025
Kind
Deep research
Level
Expert
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Strongest model tested
Qwen3.7-Plus #78 on the capability index
Last in a launch post
June 2026, dropped before it saturated
Last model added
June 2026
Scores refreshed
November 12th, 2026
3
Models scored in lab cards
3
Launch posts list it, from 3 labs

Step 3.7 Flash leads with 59% among models released by June 2026

Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.

Qwen2.5-VL-72B leads with 11.5% across all independent runs

Best score per model across the official leaderboard, Epoch AI’s own runs and any board Epoch AI copied. Each bar and row names its source.

Step 3.7 Flash leads with 59%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

The newest model with a score here came out in June 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 59.0 Step 3.7 Flash
  2. 51.9 GLM-5V-Turbo
  3. 51.1 Qwen3.7-Plus
  1. 1 Step 3.7 Flash 59%
  2. 2 GLM-5V-Turbo 51.9%
  3. 3 Qwen3.7-Plus 51.1%

Source: Labs’ model cards

Source: Official BrowseComp-VL leaderboard

Sources: Labs’ model cards and official BrowseComp-VL leaderboard

#ModelScoreSettingDateSource
1 Step 3.7 Flash 59% May 2026 Lab model card
2 GLM-5V-Turbo 51.9% Apr 2026 Lab model card
3 Qwen3.7-Plus 51.1% Jun 2026 Lab model card
#ModelScoreSettingDateSource
1 Qwen2.5-VL-72B 11.5% RAG Workflow Jan 2025 Official leaderboard
2 Claude 3.7 Sonnet 11.2% RAG Workflow Feb 2025 Official leaderboard
3 Qwen2.5-VL-32B 7.20% RAG Workflow Mar 2025 Official leaderboard
4 Qwen2.5-VL-7B 2.70% RAG Workflow Jan 2025 Official leaderboard
#ModelScoreSettingDateSource
1 Step 3.7 Flash 59% May 2026 Lab model card
2 GLM-5V-Turbo 51.9% Apr 2026 Lab model card
3 Qwen3.7-Plus 51.1% Jun 2026 Lab model card
4 Qwen2.5-VL-72B 11.5% RAG Workflow Jan 2025 Official leaderboard
5 Claude 3.7 Sonnet 11.2% RAG Workflow Feb 2025 Official leaderboard
6 Qwen2.5-VL-32B 7.20% RAG Workflow Mar 2025 Official leaderboard
7 Qwen2.5-VL-7B 2.70% RAG Workflow Jan 2025 Official leaderboard

The benchmark's own leaderboard has 8 entries, 4 of them models we track. Switch the scores above to rank them together.

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
Qwen2.5-VL-72B
Direct Infe… RAG Workflow
11.5%
Claude 3.7 Sonnet
Direct Infe… RAG Workflow
11.2%
Qwen2.5-VL-32B
Direct Infe… RAG Workflow
7.20%
Qwen2.5-VL-7B
Direct Infe… RAG Workflow
2.70%

Source: Official BrowseComp-VL leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen2.5-VL-72B RAG Workflow 11.5%
Claude 3.7 Sonnet RAG Workflow 11.2%
Qwen2.5-VL-32B RAG Workflow 7.20%
Qwen2.5-VL-72B Direct Inference 7.10%
Claude 3.7 Sonnet Direct Inference 7.10%
Qwen2.5-VL-7B RAG Workflow 2.70%
Qwen2.5-VL-32B Direct Inference 2.40%
Qwen2.5-VL-7B Direct Inference 0.50%

2 models have held number one in labs’ own numbers since its release in August 2025 Number one over time 3 models have held number one with claims included since its release in August 2025

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and what labs claim in their cards.

Qwen Z.ai StepFun
Own leaderboard Not tested since Aug 2025
Qwen2.5-VL-72B 11.5%
Lab claims Not tested since Jun 2026
Step 3.7 Flash 59%

Sources: Official BrowseComp-VL leaderboard and labs’ model cards

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Step 3.7 Flash GLM-5V-Turbo Qwen2.5-VL-72B 0% 20% 40% 60% 80% 100% 2025 2026 Near the ceiling (85%) Step 3.7 Flash GLM-5V-Turbo

Sources: Official BrowseComp-VL leaderboard and labs’ model cards

Labs headline it less often now, last in June 2026

Launch posts from every lab that has listed it, one tick per post. Tall colored ticks include it, and short gray ones leave it out.

Z.ai
StepFun
Qwen
DateLaunch postScoreAs the post writes it
Jun 1, 2026 Qwen3.7-Plus 51.1 BC-VL (with search)
May 29, 2026 Step 3.7 Flash 59 BC-VL
Apr 1, 2026 GLM-5V-Turbo 51.9 BrowseComp-VL 51.9

More in Tool use

All 144 in Tool use
  • AutomationBench In use Agent completes end-to-end business workflows inside a simulated company exposing dozens of REST APIs. Graded on resulting state. Gemini 4 Carbon 53.9%
  • MCP Atlas In use Multi-step workflows over real MCP servers. Scores tool-use competency on realistic tasks. Gemini 4 Argon 90.7%
  • BFCL V4 In use Models call functions and tools, including multi-turn agentic tasks with web search and memory. Checked against expected calls and outcomes. Qwen3.7-Max 75%
  • τ³-bench Banking In use Agents act as banking customer-service reps over many turns with a simulated user, following policy. Pass^1 on the final database state. Qwen 3.8 Max 55.1%
  • Toolathlon-Verified In use 108 long-horizon tasks across 32 apps through about 600 MCP tools. Execution-based checkers inspect final environment state. GLM-5.3-Flash 78.4%
  • DeepSearchQA In use Questions that need multi-step search and synthesis. Final answers graded against references. Kimi K3 95%