Benchmark Almanac

BrowseComp

Claude Fable 5 93.3% leads since Jun 2026 In use

Hard-to-find facts that need persistent web browsing. Short verifiable answers graded for exact correctness.

Questions ship encrypted in simple-evals. Context-management settings (compaction at 300K vs full 1M) change scores by several points.

In the authors’ words

We present BrowseComp, a simple yet challenging benchmark for measuring the ability for agents to browse the web. BrowseComp comprises 1,266 questions that require persistently navigating the internet in search of hard-to-find, entangled information. Despite the difficulty of the questions, BrowseComp is simple and easy-to-use, as predicted answers are short and easily verifiable against reference answers. BrowseComp for browsing agents can be seen as analogous to how programming competitions are an incomplete but useful benchmark for coding agents. While BrowseComp sidesteps challenges of a true user query distribution, like generating long answers or resolving ambiguity, it measures the important core capability of exercising persistence and creativity in finding information. BrowseComp can be found at https://github.com/openai/simple-evals.

Abstract of BrowseComp: A Simple Yet Challenging Benchmark for Browsing Agents (arXiv:2504.12516, CC0)
Made by OpenAI Released Apr 2025 Web browsing All tasks public Last in a launch post Oct 2026 Last model added Oct 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
openai/simple-evals 4,690 stars Paper
77
Models scored in lab cards
73
Launch posts list it, from 23 labs
34
System cards report it, from 13 labs

Claude Fable 5 leads with 93.3%

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.

77 models. Scroll sideways for the rest.

  1. 93.3 Claude Fable 5
  2. 92.2 GPT-5.6 Sol
  3. 92.1 Kimi K3.1 Pro
  4. 91.7 Ling-3.1-flash
  5. 91.5 GPT-6 Astra
  6. 91.2 Kimi K3
  7. 90.8 Claude Opus 5
  8. 90.1 GPT-5.5
  9. 90.1 GPT-5.5 Pro
  10. 89.3 GPT-5.4 Pro
  11. 88.7 Step 5 Preview
  12. 88.5 Claude Opus 4.8
  13. 87.5 GPT-5.6 Terra
  14. 86.9 Claude Mythos Preview
  15. 86.6 Claude Sonnet 5
  16. 86.2 Seed2.1 Pro
  17. 85.9 Gemini 3.1 Pro
  18. 84.5 Mistral Large 4 (Le Chonk, open weights)
  19. 84.2 Tencent Hy3
  20. 84.0 Claude Opus 4.6
  21. 83.5 MiniMax-M3
  22. 83.4 DeepSeek-V4-Pro
  23. 83.3 GPT-5.6 Luna
  24. 83.2 Kimi K2.6
  25. 82.7 GPT-5.4
  26. 79.9 LongCat-2.0
  27. 79.3 GLM-5.1
  28. 79.3 Claude Opus 4.7
  29. 78.9 Nemotron 3.5 Super VL
  30. 78.6 Qwen3.5 397B-A17B
  31. 77.9 GPT-5.2 Pro
  32. 77.8 MiniMax-M2.7
  33. 77.4 Inkling-Small
  34. 77.4 Beam 1.0
  35. 77.3 Seed2.0 Pro
  36. 77.1 Inkling
  37. 76.3 MiniMax-M2.5
  38. 75.9 GLM-5
  39. 75.8 Step 3.7 Flash
  40. 74.9 Kimi K2.5
  41. 74.7 Claude Sonnet 4.6
  42. 73.2 DeepSeek-V4-Flash
  43. 73.1 LongCat-Flash-Thinking-2601
  44. 69.0 Step 3.5 Flash
  45. 68.9 ChatGPT agent
  46. 67.6 DeepSeek-V3.2
  47. 67.6 Seed1.8
  48. 67.5 GLM-4.7
  49. 67.1 Tencent Hy3 preview
  50. 66.8 Nemotron 3.5 Bolt 74B-A9B
  51. 64.0 Seed2.0 Lite (0428)
  52. 63.8 Qwen3.5-122B-A10B
  53. 62.0 MiniMax-M2.1
  54. 60.2 Kimi K2 Thinking
  55. 58.3 MiMo-V2-Flash
  56. 54.9 GPT-5
  57. 53.1 Spark-X2.6-4B
  58. 49.7 o3
  59. 49.5 Sarvam 105B
  60. 49.2 Solar Pro 4
  61. 48.6 Mistral Medium 3.5
  62. 45.1 GLM-4.6
  63. 44.9 Grok 4 Fast
  64. 44.4 Nemotron 3 Ultra
  65. 44.0 MiniMax-M2
  66. 42.8 glm-4.7-flash
  67. 40.1 DeepSeek-V3.2-Exp
  68. 39.5 Nemotron 3.5 Lightning 30B-A3B
  69. 38.5 DeepSeek-V3.1-Terminus
  70. 35.5 Sarvam 30B
  71. 31.4 K-EXAONE
  72. 31.3 Nemotron 3 Super
  73. 30.0 DeepSeek-V3.1
  74. 29.6 UI-TARS-2
  75. 28.3 o4-mini
  76. 26.4 GLM-4.5
  77. 21.3 GLM-4.5-Air
  1. 1 Claude Fable 5 93.3%
  2. 2 GPT-5.6 Sol 92.2%
  3. 3 Kimi K3.1 Pro 92.1%
  4. 4 Ling-3.1-flash 91.7%
  5. 5 GPT-6 Astra 91.5%
  6. 6 Kimi K3 91.2%
  7. 7 Claude Opus 5 90.8%
  8. 8 GPT-5.5 90.1%
  9. 9 GPT-5.5 Pro 90.1%
  10. 10 GPT-5.4 Pro 89.3%
  11. 11 Step 5 Preview 88.7%
  12. 12 Claude Opus 4.8 88.5%
  13. 13 GPT-5.6 Terra 87.5%
  14. 14 Claude Mythos Preview 86.9%
  15. 15 Claude Sonnet 5 86.6%
  16. 16 Seed2.1 Pro 86.2%
  17. 17 Gemini 3.1 Pro 85.9%
  18. 18 Mistral Large 4 (Le Chonk, open weights) 84.5%
  19. 19 Tencent Hy3 84.2%
  20. 20 Claude Opus 4.6 84%
  21. 21 MiniMax-M3 83.5%
  22. 22 DeepSeek-V4-Pro 83.4%
  23. 23 GPT-5.6 Luna 83.3%
  24. 24 Kimi K2.6 83.2%
  25. 25 GPT-5.4 82.7%
  26. 26 LongCat-2.0 79.9%
  27. 27 GLM-5.1 79.3%
  28. 28 Claude Opus 4.7 79.3%
  29. 29 Nemotron 3.5 Super VL 78.9%
  30. 30 Qwen3.5 397B-A17B 78.6%
  31. 31 GPT-5.2 Pro 77.9%
  32. 32 MiniMax-M2.7 77.8%
  33. 33 Inkling-Small 77.4%
  34. 34 Beam 1.0 77.4%
  35. 35 Seed2.0 Pro 77.3%
  36. 36 Inkling 77.1%
  37. 37 MiniMax-M2.5 76.3%
  38. 38 GLM-5 75.9%
  39. 39 Step 3.7 Flash 75.8%
  40. 40 Kimi K2.5 74.9%
  41. 41 Claude Sonnet 4.6 74.7%
  42. 42 DeepSeek-V4-Flash 73.2%
  43. 43 LongCat-Flash-Thinking-2601 73.1%
  44. 44 Step 3.5 Flash 69%
  45. 45 ChatGPT agent 68.9%
  46. 46 DeepSeek-V3.2 67.6%
  47. 47 Seed1.8 67.6%
  48. 48 GLM-4.7 67.5%
  49. 49 Tencent Hy3 preview 67.1%
  50. 50 Nemotron 3.5 Bolt 74B-A9B 66.8%
  51. 51 Seed2.0 Lite (0428) 64%
  52. 52 Qwen3.5-122B-A10B 63.8%
  53. 53 MiniMax-M2.1 62%
  54. 54 Kimi K2 Thinking 60.2%
  55. 55 MiMo-V2-Flash 58.3%
  56. 56 GPT-5 54.9%
  57. 57 Spark-X2.6-4B 53.1%
  58. 58 o3 49.7%
  59. 59 Sarvam 105B 49.5%
  60. 60 Solar Pro 4 49.2%
  61. 61 Mistral Medium 3.5 48.6%
  62. 62 GLM-4.6 45.1%
  63. 63 Grok 4 Fast 44.9%
  64. 64 Nemotron 3 Ultra 44.4%
  65. 65 MiniMax-M2 44%
  66. 66 glm-4.7-flash 42.8%
  67. 67 DeepSeek-V3.2-Exp 40.1%
  68. 68 Nemotron 3.5 Lightning 30B-A3B 39.5%
  69. 69 DeepSeek-V3.1-Terminus 38.5%
  70. 70 Sarvam 30B 35.5%
  71. 71 K-EXAONE 31.4%
  72. 72 Nemotron 3 Super 31.3%
  73. 73 DeepSeek-V3.1 30%
  74. 74 UI-TARS-2 29.6%
  75. 75 o4-mini 28.3%
  76. 76 GLM-4.5 26.4%
  77. 77 GLM-4.5-Air 21.3%
#ModelScoreSettingDateSource
1 Claude Fable 5 93.3% Jun 2026 Model card
2 GPT-5.6 Sol 92.2% Jul 2026 Model card
3 Kimi K3.1 Pro 92.1% Oct 2026 Model card
4 Ling-3.1-flash 91.7% Oct 2026 Model card
5 GPT-6 Astra 91.5% Sep 2026 Model card
6 Kimi K3 91.2% Jul 2026 Model card
7 Claude Opus 5 90.8% Jul 2026 Model card
8 GPT-5.5 90.1% Apr 2026 Model card
9 GPT-5.5 Pro 90.1% Apr 2026 Model card
10 GPT-5.4 Pro 89.3% Mar 2026 Model card
11 Step 5 Preview 88.7% Sep 2026 Model card
12 Claude Opus 4.8 88.5% May 2026 Model card
13 GPT-5.6 Terra 87.5% Jul 2026 Model card
14 Claude Mythos Preview 86.9% Apr 2026 Model card
15 Claude Sonnet 5 86.6% Jun 2026 Model card
16 Seed2.1 Pro 86.2% Jun 2026 Model card
17 Gemini 3.1 Pro 85.9% Feb 2026 Model card
18 Mistral Large 4 (Le Chonk, open weights) 84.5% Oct 2026 Model card
19 Tencent Hy3 84.2% Jul 2026 Model card
20 Claude Opus 4.6 84% Feb 2026 Model card
21 MiniMax-M3 83.5% Jun 2026 Model card
22 DeepSeek-V4-Pro 83.4% Apr 2026 Model card
23 GPT-5.6 Luna 83.3% Jul 2026 Model card
24 Kimi K2.6 83.2% Apr 2026 Model card
25 GPT-5.4 82.7% Mar 2026 Model card
26 LongCat-2.0 79.9% Jul 2026 Model card
27 GLM-5.1 79.3% Apr 2026 Model card
28 Claude Opus 4.7 79.3% Apr 2026 Model card
29 Nemotron 3.5 Super VL 78.9% Oct 2026 Model card
30 Qwen3.5 397B-A17B 78.6% Feb 2026 Model card
31 GPT-5.2 Pro 77.9% Dec 2025 Model card
32 MiniMax-M2.7 77.8% May 2026 Model card
33 Inkling-Small 77.4% Jul 2026 Model card
34 Beam 1.0 77.4% Oct 2026 Model card
35 Seed2.0 Pro 77.3% Feb 2026 Model card
36 Inkling 77.1% Jul 2026 Model card
37 MiniMax-M2.5 76.3% Feb 2026 Model card
38 GLM-5 75.9% Feb 2026 Model card
39 Step 3.7 Flash 75.8% May 2026 Model card
40 Kimi K2.5 74.9% Jan 2026 Model card
41 Claude Sonnet 4.6 74.7% Feb 2026 Model card
42 DeepSeek-V4-Flash 73.2% Apr 2026 Model card
43 LongCat-Flash-Thinking-2601 73.1% Jan 2026 Model card
44 Step 3.5 Flash 69% Feb 2026 Model card
45 ChatGPT agent 68.9% Jul 2025 Model card
46 DeepSeek-V3.2 67.6% Dec 2025 Model card
47 Seed1.8 67.6% Dec 2025 Model card
48 GLM-4.7 67.5% Dec 2025 Model card
49 Tencent Hy3 preview 67.1% Apr 2026 Model card
50 Nemotron 3.5 Bolt 74B-A9B 66.8% Oct 2026 Model card
51 Seed2.0 Lite (0428) 64% Apr 2026 Model card
52 Qwen3.5-122B-A10B 63.8% Feb 2026 Model card
53 MiniMax-M2.1 62% Dec 2025 Model card
54 Kimi K2 Thinking 60.2% Nov 2025 Model card
55 MiMo-V2-Flash 58.3% Dec 2025 Model card
56 GPT-5 54.9% Aug 2025 Model card
57 Spark-X2.6-4B 53.1% Oct 2026 Model card
58 o3 49.7% Apr 2025 Model card
59 Sarvam 105B 49.5% Mar 2026 Model card
60 Solar Pro 4 49.2% Aug 2026 Model card
61 Mistral Medium 3.5 48.6% May 2026 Model card
62 GLM-4.6 45.1% Sep 2025 Model card
63 Grok 4 Fast 44.9% Sep 2025 Model card
64 Nemotron 3 Ultra 44.4% Jun 2026 Model card
65 MiniMax-M2 44% Oct 2025 Model card
66 glm-4.7-flash 42.8% Jan 2026 Model card
67 DeepSeek-V3.2-Exp 40.1% Sep 2025 Model card
68 Nemotron 3.5 Lightning 30B-A3B 39.5% Oct 2026 Model card
69 DeepSeek-V3.1-Terminus 38.5% Sep 2025 Model card
70 Sarvam 30B 35.5% Mar 2026 Model card
71 K-EXAONE 31.4% Dec 2025 Model card
72 Nemotron 3 Super 31.3% Mar 2026 Model card
73 DeepSeek-V3.1 30% Aug 2025 Model card
74 UI-TARS-2 29.6% Sep 2025 Model card
75 o4-mini 28.3% Apr 2025 Model card
76 GLM-4.5 26.4% Jul 2025 Model card
77 GLM-4.5-Air 21.3% Aug 2025 Model card

8 models have held number one since its release in Apr 2025

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

OpenAI Anthropic Google DeepMind
Lab claims Passed 85% in Feb 2026, by Gemini 3.1 Pro
Claude Fable 5 93.3%

The best score passed 85% in Feb 2026

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

Reported by the lab in its model card Best so far
0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Claude Fable 5 Gemini 3.1 Pro GPT-5.2 Pro ChatGPT agent o3 0% 20% 40% 60% 80% 100% 2025 2026 2027 Near the ceiling (85%) Claude Fable 5 Gemini 3.1 Pro

14 labs put it in a launch post in the last six months

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.

OpenAI
Z.ai
DeepSeek
ByteDance Seed
xAI
MiniMax
Moonshot AI
Xiaomi Corp
LG AI Research
Meituan
StepFun
Anthropic
Qwen
Google DeepMind
Sarvam AI
NVIDIA
Tencent
Mistral AI
Thinking Machines
Upstage
Reflection
Ant Group
iFlytek
DateLaunch postScoreAs the post writes it
29 Oct 2026 Mistral Large 4 (Le Chonk, open weights) 84.5 Web browsing BrowseComp
24 Oct 2026 Kimi K3.1 Pro 92.1 BrowseComp
22 Oct 2026 Spark-X2.6-4B 53.1 BrowseComp
15 Oct 2026 Nemotron 3.5 Bolt 74B-A9B 66.8 BrowseComp
15 Oct 2026 Nemotron 3.5 Super VL 78 BrowseComp (78.91 in the technical report)
13 Oct 2026 Ling-3.1-flash 91.7 Browsecomp
5 Oct 2026 Beam 1.0 77.4 BrowseComp w/ context management
20 Sep 2026 Step 5 Preview 88.7 BrowseComp
3 Sep 2026 GPT-6 Astra 91.5 BrowseComp
11 Aug 2026 Nemotron 3.5 Lightning 30B-A3B 37 BrowseComp
11 Aug 2026 Solar Pro 4 49.2 BrowseComp* (in-house)
30 Jul 2026 Inkling-Small 77.4 BrowseComp with context management
24 Jul 2026 Claude Opus 5 90.8 Agentic search BrowseComp
16 Jul 2026 Kimi K3 91.2 BrowseComp
15 Jul 2026 Inkling 77.1 BrowseComp w/ ctx management
9 Jul 2026 GPT-5.6 Luna 83.3 BrowseComp
9 Jul 2026 GPT-5.6 Terra 87.5 BrowseComp
9 Jul 2026 GPT-5.6 Sol 92.2 BrowseComp (Sol Ultra, 4 agents)
5 Jul 2026 LongCat-2.0 79.9 BrowseComp
2 Jul 2026 Tencent Hy3 84.2 BrowseComp
30 Jun 2026 Claude Sonnet 5 86.6 BrowseComp (multi agent)
9 Jun 2026 Claude Fable 5 93.3 BrowseComp (multi-agent)
4 Jun 2026 Nemotron 3 Ultra 44.4 BrowseComp
1 Jun 2026 MiniMax-M3 83.5 BrowseComp
29 May 2026 Step 3.7 Flash 75.8 BrowseComp (acc)
28 May 2026 Claude Opus 4.8 88.5 BrowseComp (multi-agent)
22 May 2026 Mistral Medium 3.5 48.6 BrowseComp**** (self-reported, context management, discard-all at 100k tokens)
28 Apr 2026 Seed2.0 Lite (0428) 64 BrowseComp
24 Apr 2026 DeepSeek-V4-Pro 83.4 BrowseComp (Pass@1)
23 Apr 2026 GPT-5.5 Pro 90.1 BrowseComp
23 Apr 2026 Tencent Hy3 preview 67.1 BrowseComp
23 Apr 2026 GPT-5.5 90.1 BrowseComp (GPT-5.5 Pro)
20 Apr 2026 Kimi K2.6 83.2 BrowseComp
16 Apr 2026 Claude Opus 4.7 79.3 Agentic search BrowseComp
7 Apr 2026 GLM-5.1 79.3 BrowseComp (w/ Context Manage)
7 Apr 2026 Claude Mythos Preview 86.9 BrowseComp (Glasswing page)
11 Mar 2026 Nemotron 3 Super 31.3 BrowseComp with Search
6 Mar 2026 Sarvam 30B 35.5 BrowseComp
6 Mar 2026 Sarvam 105B 49.5 BrowseComp
5 Mar 2026 GPT-5.4 Pro 89.3 BrowseComp
5 Mar 2026 GPT-5.4 82.7 BrowseComp
24 Feb 2026 Qwen3.5-122B-A10B 63.8 Browsecomp (context folding)
19 Feb 2026 Gemini 3.1 Pro 85.9 BrowseComp (Search + Python + Browse)
17 Feb 2026 Claude Sonnet 4.6 74.7 Agentic search BrowseComp
16 Feb 2026 Qwen3.5 397B-A17B 78.6 BrowseComp (69.0/78.6)
14 Feb 2026 Seed2.0 Pro 77.3 BrowseComp
12 Feb 2026 MiniMax-M2.5 76.3 BrowseComp (w/ctx)
11 Feb 2026 GLM-5 75.9 BrowseComp (w/ Context Manage)
5 Feb 2026 Claude Opus 4.6 84 Agentic search BrowseComp
2 Feb 2026 Step 3.5 Flash 69 Best of BrowseComp 51.6 / BrowseComp (w/ Context Manager) 69.0
27 Jan 2026 Kimi K2.5 74.9 BrowseComp (w/ctx manage)
19 Jan 2026 glm-4.7-flash 42.8 BrowseComp
14 Jan 2026 LongCat-Flash-Thinking-2601 73.1 Best of BrowseComp (Pass@1) 56.6 / with context management 73.1
31 Dec 2025 K-EXAONE 31.4 BrowseComp
23 Dec 2025 MiniMax-M2.1 62 BrowseComp (context management)
22 Dec 2025 GLM-4.7 67.5 BrowseComp (w/ Context Manage)
18 Dec 2025 Seed1.8 67.6 BrowseComp-en
16 Dec 2025 MiMo-V2-Flash 58.3 Best of BrowseComp 45.4 / BrowseComp (w/ Context Manage) 58.3
11 Dec 2025 GPT-5.2 Pro 77.9 BrowseComp
1 Dec 2025 DeepSeek-V3.2 67.6 BrowseComp (Pass@1)
6 Nov 2025 Kimi K2 Thinking 60.2 BrowseComp, w/ tools
27 Oct 2025 MiniMax-M2 44 BrowseComp
30 Sep 2025 GLM-4.6 45.1 BrowseComp
29 Sep 2025 DeepSeek-V3.2-Exp 40.1 BrowseComp
22 Sep 2025 DeepSeek-V3.1-Terminus 38.5 BrowseComp
19 Sep 2025 Grok 4 Fast 44.9 BrowseComp pass@1
2 Sep 2025 UI-TARS-2 29.6 BC-en 7.0 GUI-only (29.6 with GUI-SDK)
21 Aug 2025 DeepSeek-V3.1 30 BrowseComp
7 Aug 2025 GPT-5 54.9 BrowseComp Agentic search & browsing
28 Jul 2025 GLM-4.5 26.4 BrowseComp
17 Jul 2025 ChatGPT agent 68.9 BrowseComp (Agentic Browsing)
16 Apr 2025 o4-mini 28.3 BrowseComp (with python + browsing)
16 Apr 2025 o3 49.7 BrowseComp Agentic Browsing, o3 with python + browsing

34 system cards report it, from 13 labs

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
15 Oct 2026 Nemotron 3.5 Super VL technical report 78.9 BrowseComp, NeMo Gym harness with Exa search (Table 10). OpenCode harness gives 72.35 Capabilities
15 Oct 2026 Nemotron 3.5 Lightning 30B-A3B technical report 39.5 BrowseComp, BF16 acc. (Table 22) Capabilities
15 Oct 2026 Nemotron 3.5 Bolt 74B-A9B technical report 66.8 BrowseComp, Exa web search, released 24GB checkpoint (Table 7 and Section 8). BF16 70.00 Capabilities
30 Jul 2026 Inkling-Small model card 77.4 BrowseComp with context management (5. Evaluations table) Capabilities
27 Jul 2026 Kimi K3 technical report 91.2 BrowseComp with context compaction at 300K tokens, 90.4 with full 1M context and no context management (Table 2, 6.1.3) Capabilities
24 Jul 2026 Claude Opus 5 system card 90.8 BrowseComp, single agent with web search, fetch, programmatic tool calling, code execution and compaction (Table 8.1.A, 8.10.2). Multi-agent 10-agent team 93.6% on a pre-release configuration (8.11.1) Capabilities
15 Jul 2026 Inkling model card 77.1 BrowseComp w/ ctx management (5. Evaluations table, effort 0.99) Capabilities
30 Jun 2026 Claude Sonnet 5 system card 84.7 BrowseComp single agent, max effort, 10M-token limit with compaction (multi agent 86.6%) (Table 8.1.A, 8.9.2) Capabilities
23 Jun 2026 Seed2.1 Pro model card 86.2 BrowseComp (with Search) (Table 11) Capabilities
9 Jun 2026 Nemotron 3 Ultra technical report 44.4 BrowseComp (Table 10) Capabilities
28 May 2026 Claude Opus 4.8 system card 84.3 BrowseComp single-agent, max effort, 10M-token limit with compaction at 200k (multi-agent 88.5% with an orchestrator and blocking subagents, five-agent team 85.4%) (Table 8.1.A, 8.10.2, 8.11.1) Capabilities
26 May 2026 MiniMax-M2 technical report 44 BrowseComp (Hugging Face official score), MiniMax-M2 bar, printed label (Figure 9) Capabilities
26 May 2026 MiniMax-M2.7 technical report 77.8 BrowseComp, WebExplorer framework with context dropping (Table 4) Capabilities
26 May 2026 MiniMax-M2.5 technical report 76.3 BrowseComp, WebExplorer framework with context dropping (Table 4) Capabilities
26 Apr 2026 DeepSeek-V4-Pro technical report 83.4 BrowseComp (Pass@1), discard-all context management, DS-V4-Pro Max (Table 6) Capabilities
26 Apr 2026 DeepSeek-V4-Flash technical report 73.2 BrowseComp (Pass@1), DeepSeek-V4-Flash Max (Table 7) Capabilities
16 Apr 2026 Claude Opus 4.7 system card 79.3 BrowseComp, thinking off at max effort, 10M token limit with compaction (Table 8.1.A, 8.8.2) Capabilities
7 Apr 2026 Claude Mythos Preview system card 86.9 BrowseComp, web search, web fetch, programmatic tool calling and code execution, adaptive thinking, max effort, 3M token limit with compaction (6.10.2); 24.0 with no thinking and no tools as a contamination check Capabilities
21 Mar 2026 Seed1.8 model card 67.6 BrowseComp-en (Table 4). 45.0 low and 55.0 medium effort (Figure 4 text) Capabilities
11 Mar 2026 Nemotron 3 Super technical report 31.3 BrowseComp with Search, internal harness with Serp API (Table 5) Capabilities
19 Feb 2026 Gemini 3.1 Pro model card 85.9 BrowseComp, Search + Python + Browse (results table, p.4) Capabilities
17 Feb 2026 GLM-5 technical report 75.9 BrowseComp (w/ Context Manage), discard-all (Table 7, Figure 1 headline). Without context management 62.0 Capabilities
17 Feb 2026 Claude Sonnet 4.6 system card 74 BrowseComp single agent, no thinking, max effort, compaction up to 10M tokens, revised 2026-03-06 from 74.72; multi-agent 82.07 (2.20.1.1, 2.20.1.3) Capabilities
14 Feb 2026 Seed2.0 Pro model card 77.3 BrowseComp, search agent (Table 11) Capabilities
11 Feb 2026 Step 3.5 Flash technical report 69 BrowseComp (w. Ctx Manage), discard-all context management, as in the abstract (Table 5). Without context management 51.6 Capabilities
5 Feb 2026 Claude Opus 4.6 system card 83.7 BrowseComp, best single-agent configuration, no thinking, max effort, compaction up to 10M tokens, revised on 2026-03-06 from 83.97 (Figure 2.21.1.A); multi-agent 86.57, revised from 86.81 (2.21.1.2) Capabilities
30 Jan 2026 Kimi K2.5 technical report 60.6 BrowseComp without context management. 74.9 with discard-all context management, 78.4 with Agent Swarm (Table 4) Capabilities
23 Jan 2026 LongCat-Flash-Thinking-2601 technical report 73.1 BrowseComp (Pass@1), with context management (Table 2). Without context management 56.6 Capabilities
6 Jan 2026 MiMo-V2-Flash technical report 58.3 BrowseComp (w/ Context Manage) 58.3, BrowseComp without it 45.4 (Table 9) Capabilities
5 Jan 2026 K-EXAONE technical report 31.4 BrowseComp, marked Non-reasoning, with summarizer and trajectory compressor (Table 3 and C.6) Capabilities
1 Dec 2025 DeepSeek-V3.2 technical report 67.6 BrowseComp (Pass@1) with discard-all context management, the final score per 4.1. 51.4 without context management, 60.2 with Summary DeepSeek-V3.2 Thinking (Table 2) Capabilities
29 Sep 2025 DeepSeek-V3.2-Exp technical report 40.1 BrowseComp (Acc.), search agent, DeepSeek-V3.2-Exp (Table 1) Capabilities
8 Aug 2025 GLM-4.5 technical report 26.4 BrowseComp (Table 3) Capabilities
8 Aug 2025 GLM-4.5-Air technical report 21.3 BrowseComp (Table 3) Capabilities

More in Computer and browser use

All 56 in Computer and browser use
  • 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. Kimi K3 74.4%
  • RecreationBench In use Agents explore a running reference app and rebuild it, across 250 tasks on five platforms. Programmatic tests and a VLM judge score the result. GPT-6 Astra 58.1%