Benchmark Almanac

BU Bench V2.1

GPT-6.1 Sol 81.3 leads since September 2026
Not in model cards

Browser agents do 200 real web tasks. A findings judge checks tool evidence, answers and files, and deterministic integrity checks run as well.

Released as v2.1 on September 25th, 2026 with an encrypted task file. Replaces the 60-task set.

Made by
Browser Use
Released
September 2026
Kind
Web browsing
Level
Everyday
Scored
Graded by a model
Answers
Agent task
Tasks
All public
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
18
Entries on its own leaderboard, 16 models we track

GPT-6.1 Sol leads with 81.3

Best score per entry from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

18 entries. Scroll sideways for the rest.

  1. 81.3 GPT-6.1 Sol
  2. 79.4 GPT-6 Sol
  3. 79.0 GPT-6.1 Terra
  4. 77.9 GPT-6 Astra
  5. 76.7 Kimi K3.1 Pro
  6. 75.0 Claude Opus 5.5
  7. 75.0 DeepSeek V4.1 Flash 1017
  8. 73.3 Grok 4.7
  9. 72.9 Browser Use
  10. 72.7 Claude Sonnet 5.5
  11. 72.0 Claude Fable 5.1
  12. 70.9 DeepSeek V4.1 Flash
  13. 69.6 GPT-5.6 Luna
  14. 69.5 Claude Haiku 5.5
  15. 69.2 GPT-6 Luna
  16. 66.8 Claude Opus 5
  17. 59.7 GLM 5.3 FlashX
  18. 50.6 MiniMax-M3.1-Flash
  1. 1 GPT-6.1 Sol high 81.3
  2. 2 GPT-6 Sol high 79.4
  3. 3 GPT-6.1 Terra high 79.0
  4. 4 GPT-6 Astra medium 77.9
  5. 5 Kimi K3.1 Pro high 76.7
  6. 6 Claude Opus 5.5 high 75.0
  7. 7 DeepSeek V4.1 Flash 1017 max 75.0
  8. 8 Grok 4.7 high 73.3
  9. 9 Browser Use BU fast 72.9
  10. 10 Claude Sonnet 5.5 high 72.7
  11. 11 Claude Fable 5.1 high 72.0
  12. 12 DeepSeek V4.1 Flash max 70.9
  13. 13 GPT-5.6 Luna xhigh 69.6
  14. 14 Claude Haiku 5.5 xhigh 69.5
  15. 15 GPT-6 Luna max 69.2
  16. 16 Claude Opus 5 66.8
  17. 17 GLM 5.3 FlashX max 59.7
  18. 18 MiniMax-M3.1-Flash medium 50.6

Source: Official BU Bench V2.1 leaderboard

#EntryScoreSettingDateSource
1 GPT-6.1 Sol 81.3 high, BrowserCode Sep 2026 Official leaderboard
2 GPT-6 Sol 79.4 high, BrowserCode Sep 2026 Official leaderboard
3 GPT-6.1 Terra 79.0 high, BrowserCode Oct 2026 Official leaderboard
4 GPT-6 Astra 77.9 medium, BrowserCode Sep 2026 Official leaderboard
5 Kimi K3.1 Pro 76.7 high, BrowserCode Oct 2026 Official leaderboard
6 Claude Opus 5.5 75.0 high, BrowserCode Sep 2026 Official leaderboard
7 DeepSeek V4.1 Flash 1017 75.0 max, BrowserCode Oct 2026 Official leaderboard
8 Grok 4.7 73.3 high, BrowserCode Sep 2026 Official leaderboard
9 Browser Use 72.9 BU fast, Browser Use Official leaderboard
10 Claude Sonnet 5.5 72.7 high, BrowserCode Sep 2026 Official leaderboard
11 Claude Fable 5.1 72.0 high, BrowserCode Sep 2026 Official leaderboard
12 DeepSeek V4.1 Flash 70.9 max, BrowserCode Sep 2026 Official leaderboard
13 GPT-5.6 Luna 69.6 xhigh, BrowserCode Jul 2026 Official leaderboard
14 Claude Haiku 5.5 69.5 xhigh, BrowserCode Oct 2026 Official leaderboard
15 GPT-6 Luna 69.2 max, BrowserCode Sep 2026 Official leaderboard
16 Claude Opus 5 66.8 BrowserCode Jul 2026 Official leaderboard
17 GLM 5.3 FlashX 59.7 max, BrowserCode Official leaderboard
18 MiniMax-M3.1-Flash 50.6 medium, BrowserCode Oct 2026 Official leaderboard

Score vs. cost per task

The line joins the runs that no other run beats on both measures. We redrew this chart from the numbers on the official leaderboard.

0 20 40 60 80 100 $0.01$0.1$1$10$100 Cost per task ($), log scale, lower is better Score (score) GPT-5.6 Luna (xhigh): Score 69.6 score, cost per task $0.28 GPT-5.6 Luna (none): Score 31 score, cost per task $0.04 Claude Opus 5: Score 66.8 score, cost per task $4.72 Claude Fable 5.1 (high): Score 72 score, cost per task $5.19 GPT-6 Astra (medium): Score 77.9 score, cost per task $10.8 DeepSeek V4.1 Flash (max): Score 70.9 score, cost per task $0.18 Grok 4.7 (high): Score 73.3 score, cost per task $3.46 Claude Opus 5.5 (high): Score 75 score, cost per task $2.35 GPT-6 Luna (max): Score 69.2 score, cost per task $0.22 GPT-6 Luna (xhigh): Score 69 score, cost per task $0.16 GPT-6 Luna (high): Score 68.9 score, cost per task $0.12 GPT-6 Luna (none): Score 37.6 score, cost per task $0.09 GPT-6 Sol (high): Score 79.4 score, cost per task $1.86 GPT-6 Sol (xhigh): Score 79.4 score, cost per task $3 GPT-6 Sol (medium): Score 74.4 score, cost per task $1.02 GPT-6 Sol (max): Score 71.5 score, cost per task $4.71 GPT-6 Sol (none): Score 63.5 score, cost per task $1.21 GPT-6 Sol (low): Score 63.3 score, cost per task $0.44 Claude Sonnet 5.5 (high): Score 72.7 score, cost per task $1.53 GPT-6.1 Sol (xhigh): Score 78.5 score, cost per task $2.23 GPT-6.1 Sol (low): Score 74.7 score, cost per task $0.75 Claude Haiku 5.5 (xhigh): Score 69.5 score, cost per task $0.46 Claude Haiku 5.5 (max): Score 65.3 score, cost per task $1.07 Claude Haiku 5.5 (high): Score 64.3 score, cost per task $0.23 Claude Haiku 5.5 (medium): Score 53.4 score, cost per task $0.09 DeepSeek V4.1 Flash 1017 (high): Score 73.6 score, cost per task $0.15 MiniMax-M3.1-Flash (medium): Score 50.6 score, cost per task $0.13 MiniMax-M3.1-Flash (high): Score 49.5 score, cost per task $0.13 MiniMax-M3.1-Flash (low): Score 48.6 score, cost per task $0.11 Kimi K3.1 Pro (high): Score 76.7 score, cost per task $1.89 GLM 5.3 FlashX (max): Score 59.7 score, cost per task $0.55. Not in our model list GPT-6 Luna (medium): Score 64.3 score, cost per task $0.08 GPT-6 Luna (low): Score 36.2 score, cost per task $0.01 GPT-6.1 Sol (high): Score 81.3 score, cost per task $1.92 GPT-6.1 Sol (medium): Score 79.6 score, cost per task $1.3 Claude Haiku 5.5 (low): Score 45.6 score, cost per task $0.03 DeepSeek V4.1 Flash 1017 (max): Score 75 score, cost per task $0.22 DeepSeek V4.1 Flash 1017 (low): Score 74 score, cost per task $0.12 GPT-6.1 Terra (high): Score 79 score, cost per task $1.26 Browser Use (BU fast): Score 72.9 score, cost per task $0.11. Not in our model list GPT-6.1 Sol (high)GPT-6.1 Sol (medium)GPT-6.1 Terra (high)DeepSeek V4.1 Flash 1017 (max)DeepSeek V4.1 Flash 1017 (low)Browser Use (BU fast)GPT-6 Luna (medium)Claude Haiku 5.5 (low)GPT-6 Luna (low) 0 20 40 60 80 100 $0.01$0.1$1$10$100 Cost per task ($), log scale, lower is better Score (score) GPT-5.6 Luna (xhigh): Score 69.6 score, cost per task $0.28 GPT-5.6 Luna (none): Score 31 score, cost per task $0.04 Claude Opus 5: Score 66.8 score, cost per task $4.72 Claude Fable 5.1 (high): Score 72 score, cost per task $5.19 GPT-6 Astra (medium): Score 77.9 score, cost per task $10.8 DeepSeek V4.1 Flash (max): Score 70.9 score, cost per task $0.18 Grok 4.7 (high): Score 73.3 score, cost per task $3.46 Claude Opus 5.5 (high): Score 75 score, cost per task $2.35 GPT-6 Luna (max): Score 69.2 score, cost per task $0.22 GPT-6 Luna (xhigh): Score 69 score, cost per task $0.16 GPT-6 Luna (high): Score 68.9 score, cost per task $0.12 GPT-6 Luna (none): Score 37.6 score, cost per task $0.09 GPT-6 Sol (high): Score 79.4 score, cost per task $1.86 GPT-6 Sol (xhigh): Score 79.4 score, cost per task $3 GPT-6 Sol (medium): Score 74.4 score, cost per task $1.02 GPT-6 Sol (max): Score 71.5 score, cost per task $4.71 GPT-6 Sol (none): Score 63.5 score, cost per task $1.21 GPT-6 Sol (low): Score 63.3 score, cost per task $0.44 Claude Sonnet 5.5 (high): Score 72.7 score, cost per task $1.53 GPT-6.1 Sol (xhigh): Score 78.5 score, cost per task $2.23 GPT-6.1 Sol (low): Score 74.7 score, cost per task $0.75 Claude Haiku 5.5 (xhigh): Score 69.5 score, cost per task $0.46 Claude Haiku 5.5 (max): Score 65.3 score, cost per task $1.07 Claude Haiku 5.5 (high): Score 64.3 score, cost per task $0.23 Claude Haiku 5.5 (medium): Score 53.4 score, cost per task $0.09 DeepSeek V4.1 Flash 1017 (high): Score 73.6 score, cost per task $0.15 MiniMax-M3.1-Flash (medium): Score 50.6 score, cost per task $0.13 MiniMax-M3.1-Flash (high): Score 49.5 score, cost per task $0.13 MiniMax-M3.1-Flash (low): Score 48.6 score, cost per task $0.11 Kimi K3.1 Pro (high): Score 76.7 score, cost per task $1.89 GLM 5.3 FlashX (max): Score 59.7 score, cost per task $0.55. Not in our model list GPT-6 Luna (medium): Score 64.3 score, cost per task $0.08 GPT-6 Luna (low): Score 36.2 score, cost per task $0.01 GPT-6.1 Sol (high): Score 81.3 score, cost per task $1.92 GPT-6.1 Sol (medium): Score 79.6 score, cost per task $1.3 Claude Haiku 5.5 (low): Score 45.6 score, cost per task $0.03 DeepSeek V4.1 Flash 1017 (max): Score 75 score, cost per task $0.22 DeepSeek V4.1 Flash 1017 (low): Score 74 score, cost per task $0.12 GPT-6.1 Terra (high): Score 79 score, cost per task $1.26 Browser Use (BU fast): Score 72.9 score, cost per task $0.11. Not in our model list GPT-6.1 Sol (high)GPT-6.1 Sol (medium)GPT-6.1 Terra (high)DeepSeek V4.1 Flash 1017 (max)DeepSeek V4.1 Flash 1017 (low)Browser Use (BU fast)GPT-6 Luna (medium)Claude Haiku 5.5 (low)GPT-6 Luna (low)
ModelSettingScoreCost per task
GPT-6.1 Sol high, BrowserCode 81.3 score $1.92
GPT-6.1 Sol medium, BrowserCode 79.6 score $1.3
GPT-6 Sol high, BrowserCode 79.4 score $1.86
GPT-6 Sol xhigh, BrowserCode 79.4 score $3
GPT-6.1 Terra high, BrowserCode 79 score $1.26
GPT-6.1 Sol xhigh, BrowserCode 78.5 score $2.23
GPT-6 Astra medium, BrowserCode 77.9 score $10.77
Kimi K3.1 Pro high, BrowserCode 76.7 score $1.89
DeepSeek V4.1 Flash 1017 max, BrowserCode 75 score $0.22
Claude Opus 5.5 high, BrowserCode 75 score $2.35
GPT-6.1 Sol low, BrowserCode 74.7 score $0.75
GPT-6 Sol medium, BrowserCode 74.4 score $1.02
DeepSeek V4.1 Flash 1017 low, BrowserCode 74 score $0.12
DeepSeek V4.1 Flash 1017 high, BrowserCode 73.6 score $0.15
Grok 4.7 high, BrowserCode 73.3 score $3.46
Browser Use BU fast, Browser Use 72.9 score $0.11
Claude Sonnet 5.5 high, BrowserCode 72.7 score $1.53
Claude Fable 5.1 high, BrowserCode 72 score $5.19
GPT-6 Sol max, BrowserCode 71.5 score $4.71
DeepSeek V4.1 Flash max, BrowserCode 70.9 score $0.18
GPT-5.6 Luna xhigh, BrowserCode 69.6 score $0.28
Claude Haiku 5.5 xhigh, BrowserCode 69.5 score $0.46
GPT-6 Luna max, BrowserCode 69.2 score $0.22
GPT-6 Luna xhigh, BrowserCode 69 score $0.16
GPT-6 Luna high, BrowserCode 68.9 score $0.12
Claude Opus 5 BrowserCode 66.8 score $4.72
Claude Haiku 5.5 max, BrowserCode 65.3 score $1.07
GPT-6 Luna medium, BrowserCode 64.3 score $0.08
Claude Haiku 5.5 high, BrowserCode 64.3 score $0.23
GPT-6 Sol none, BrowserCode 63.5 score $1.21
GPT-6 Sol low, BrowserCode 63.3 score $0.44
GLM 5.3 FlashX max, BrowserCode 59.7 score $0.55
Claude Haiku 5.5 medium, BrowserCode 53.4 score $0.09
MiniMax-M3.1-Flash medium, BrowserCode 50.6 score $0.13
MiniMax-M3.1-Flash high, BrowserCode 49.5 score $0.13
MiniMax-M3.1-Flash low, BrowserCode 48.6 score $0.11
Claude Haiku 5.5 low, BrowserCode 45.6 score $0.03
GPT-6 Luna none, BrowserCode 37.6 score $0.09
GPT-6 Luna low, BrowserCode 36.2 score $0.01
GPT-5.6 Luna none, BrowserCode 31 score $0.04

The best score, 81.3, costs $1.92 a task

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Anthropic DeepSeek Cost frontier One model at each effort level
0.00 20.0 40.0 60.0 80.0 100.0 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.1 Sol, high DeepSeek V4.1 Flash 1017, max GPT-6 Luna, medium Claude Haiku 5.5, low GPT-6 Luna, low 0.00 20.0 40.0 60.0 80.0 100.0 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale GPT-6.1 Sol, high DeepSeek V4.1 Flash 1017, max

Source: Official BU Bench V2.1 leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-6.1 Sol high BrowserCode 81.3 $1.92
GPT-6.1 Sol medium BrowserCode 79.6 $1.30
GPT-6 Sol high BrowserCode 79.4 $1.86
GPT-6 Sol xhigh BrowserCode 79.4 $3.00
GPT-6.1 Terra high BrowserCode 79.0 $1.26
GPT-6.1 Sol xhigh BrowserCode 78.5 $2.23
GPT-6 Astra medium BrowserCode 77.9 $11
Kimi K3.1 Pro high BrowserCode 76.7 $1.89
Claude Opus 5.5 high BrowserCode 75.0 $2.35
DeepSeek V4.1 Flash 1017 max BrowserCode 75.0 $0.22
GPT-6.1 Sol low BrowserCode 74.7 $0.75
GPT-6 Sol medium BrowserCode 74.4 $1.02
DeepSeek V4.1 Flash 1017 low BrowserCode 74.0 $0.12
DeepSeek V4.1 Flash 1017 high BrowserCode 73.6 $0.15
Grok 4.7 high BrowserCode 73.3 $3.46
Claude Sonnet 5.5 high BrowserCode 72.7 $1.53
Claude Fable 5.1 high BrowserCode 72.0 $5.19
GPT-6 Sol max BrowserCode 71.5 $4.71
DeepSeek V4.1 Flash max BrowserCode 70.9 $0.18
GPT-5.6 Luna xhigh BrowserCode 69.6 $0.28
Claude Haiku 5.5 xhigh BrowserCode 69.5 $0.46
GPT-6 Luna max BrowserCode 69.2 $0.22
GPT-6 Luna xhigh BrowserCode 69.0 $0.16
GPT-6 Luna high BrowserCode 68.9 $0.12
Claude Opus 5 BrowserCode 66.8 $4.72
Claude Haiku 5.5 max BrowserCode 65.3 $1.07
GPT-6 Luna medium BrowserCode 64.3 $0.08
Claude Haiku 5.5 high BrowserCode 64.3 $0.23
GPT-6 Sol none BrowserCode 63.5 $1.21
GPT-6 Sol low BrowserCode 63.3 $0.44
Claude Haiku 5.5 medium BrowserCode 53.4 $0.09
MiniMax-M3.1-Flash medium BrowserCode 50.6 $0.13
MiniMax-M3.1-Flash high BrowserCode 49.5 $0.13
MiniMax-M3.1-Flash low BrowserCode 48.6 $0.11
Claude Haiku 5.5 low BrowserCode 45.6 $0.03
GPT-6 Luna none BrowserCode 37.6 $0.09
GPT-6 Luna low BrowserCode 36.2 $0.01
GPT-5.6 Luna none BrowserCode 31.0 $0.04

4 models have held number one on the official leaderboard since its release in September 2026

Lab claims count from the day the card came out, independent runs from the model’s release.

Anthropic OpenAI
Own leaderboard 4 leaders
GPT-6.1 Sol 81.3

Source: Official BU Bench V2.1 leaderboard

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
20.0 40.0 60.0 80.0 100.0 2027 20.0 40.0 60.0 80.0 100.0 2027

Source: Official BU Bench V2.1 leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

Versions

Replaces BU Bench V2

More in Computer and browser use

All 69 in Computer and browser use
  • BrowseComp In use Hard-to-find facts that need persistent web browsing. Short verifiable answers graded for exact correctness. Claude Fable 5 93.3%
  • ScreenSpot-Pro In use GUI grounding on high-resolution professional-software screenshots. The model must locate the target UI element. GPT-6 Astra 92.7%
  • AndroidWorld In use Agents complete 116 tasks across 20 real Android apps in an emulator. Each task checks the device's system state to judge success. GPT-6 Astra 100%
  • OSWorld 2.1 In use Agents complete 108 long-horizon workflows across 31 self-hosted websites and desktop apps, averaging 318 tool calls. Scored by binary completion and checkpoint partial credit. Claude Opus 5.5 81.8%
  • WebArena-Verified In use Agents complete 812 audited WebArena tasks on self-hosted websites. Deterministic evaluators check the agent's answer and captured network traces. Muse Spark 1.1 69%
  • MobileWorld In use Agents complete 201 tasks across 20 mobile apps, including tasks that need user clarification or MCP tools. Backend checks verify success. GPT-5 55.3%