Examenos

88.4% · raw 88.4 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.280, max_seq_length 512K; average of three runs. Exact label “Terminal Bench v2.1”. Self-reported; run date unpublished.

38.5% · raw 38.5 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.280, max_seq_length 512K. Exact label “Terminal Bench v4.0”. Self-reported; run date unpublished.

25.5% · raw 25.5 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.280, max_seq_length 512K. Source label “Almost Resolved”. Exact label “Program Bench Almost Resolved”. Self-reported; run date unpublished.

64.6% · raw 64.6 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. mini-swe-agent, max_seq_length 1M, average of three runs; version of mini-swe-agent unstated. Exact label “DeepSWE v1.1”. Self-reported; run date unpublished.

82.1% · raw 82.1 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Harness and judge unstated. Exact label “PaperBench”. Self-reported; run date unpublished.

58.3% · raw 58.3 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.116, Claude Opus 4.5 judge by majority of three judgments, average of three runs. Exact label “SWE Atlas - Codebase QnA”. Self-reported; run date unpublished.

55.2% · raw 55.2 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Harness unstated; the setup footnote covers only Codebase QnA and Test Writing. Exact label “SWE Atlas - Refactoring”. Self-reported; run date unpublished.

47.8% · raw 47.8 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.116, Claude Opus 4.5 judge by majority of three judgments, average of three runs. Exact label “SWE Atlas - Test Writing”. Self-reported; run date unpublished.

60.5% · raw 60.5 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.280, max_seq_length 512K. Exact label “NL2Repo”. Self-reported; run date unpublished.

73.7% · raw 73.7 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Internal benchmark. Exact label “LongCat AppDev Internal”. Self-reported; run date unpublished.

91.9% · raw 91.9 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Custom harness with Verify and Discard-All (threshold 256K), up to 1,024 rounds, tools search and browse_with_goal only. Exact label “BrowseComp”. Self-reported; run date unpublished.

51.8% · raw 51.8 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. All 600 public tasks of v1.0.6 with the official evaluation code. Exact label “AutomationBench v1.0.6”. Self-reported; run date unpublished.

63.1% · raw 63.1 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Text-only subset, Claude Code v2.1.116, max generation 256K. Exact label “HLE w/ tools”. Self-reported; run date unpublished.

38.5% · raw 38.5 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Population and harness unstated. Exact label “ALE v1.1”. Self-reported; run date unpublished.

81% · raw 81 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Harness and judge unstated. Exact label “ResearchRubrics”. Self-reported; run date unpublished.

83.1% · raw 83.1 %

Headline · max effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Effort max, temperature 0.7, top_p 0.95, 256K context. Exact label “MCP Atlas”. Self-reported; run date unpublished.

65.4% · raw 65.4 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. JobBench Main 65, Claude Code v2.1.280, official chained-rubric judging with gemini-3.6-flash. Exact label “JobBench”. Self-reported; run date unpublished.

63.3% · raw 63.3 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. v1.1, 87 tasks, Codex harness, top_k -1. Exact label “SkillsBench”. Self-reported; run date unpublished.

84.8% · raw 84.8 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Rubric score, not Elo. Exact label “GDPVal-AA rubrics”. Self-reported; run date unpublished.

77% · raw 77 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Split and judge unstated. Exact label “FrontierScience”. Self-reported; run date unpublished.

34.9% · raw 34.9 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.116, official protocol; Visualization subset judged by gemini-3.6-flash. Exact label “SpreadsheetBench 2”. Self-reported; run date unpublished.

63.4% · raw 63.4 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Harness unstated. Exact label “FinSearchComp T3”. Self-reported; run date unpublished.

78.3% · raw 78.3 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Pass³; task population unstated, so not mapped to any 157-task General setup. Exact label “Claw-Eval Pass^3”. Self-reported; run date unpublished.

83.3% · raw 83.3 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. LongCat’s own benchmark. Exact label “LongCatClawBench FORTE”. Self-reported; run date unpublished.

55.4% · raw 55.4 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Harness and judge unstated. Exact label “3DCodeBench Text→3D”. Self-reported; run date unpublished.

78.6% · raw 78.6 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Claude Code v2.1.7, top_k 50, sequence length 256K. Exact label “Vision2Web”. Self-reported; run date unpublished.

90.2% · raw 90.2 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. With tools, max generation 131,972 tokens. Exact label “CharXiv-RQ w/ tools”. Self-reported; run date unpublished.

96.5% · raw 96.5 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. With tools, max generation 131,972 tokens. Exact label “MathVision w/ tools”. Self-reported; run date unpublished.

85.2% · raw 85.2 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. At most 2,048 frames at 1 fps, 256 tokens per frame; subtitle use unstated. Exact label “VideoMME”. Self-reported; run date unpublished.

77.4% · raw 77.4 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. Harness and judge unstated. Exact label “PosterBench”. Self-reported; run date unpublished.

72.1% · raw 72.1 %

Headline · unknown effort · Own vendor

Source/record date: 2026-11-01

LongCat run with thinking on, temperature 1.0, top_p 0.95 unless stated. LongCat’s own benchmark. Exact label “LongCatClawBench-MM FORTE-MM”. Self-reported; run date unpublished.