Knowledge & reasoning

GraphWalks BFS 256K–1M leaderboard

Name: benchgap: LLM benchmark scores, measured and estimated
Creator: benchgap
License: https://creativecommons.org/licenses/by-nc/4.0/

As of 2026-10-29, the highest measured score on GraphWalks BFS 256K–1M is 96.2% by Gemini 4 Carbon. 29 more models have estimated scores, calibrated from the benchmarks they were measured on.

Measured scores: benchlm.ai.

#	Model	Score	Source
1	Gemini 4 Carbon	96.2%	measured
2	Gemini 4 Argon	84.2%	measured
3	Muse Wonder Full	72.0%	estimated ± 14.8 pp, low confidence
4	GPT-5.6 Sol	72.0%	estimated ± 14.8 pp, low confidence
5	GPT-6.1 Sol	72.0%	estimated ± 14.8 pp, low confidence
6	Claude Sonnet 5.5	72.0%	estimated ± 14.8 pp, low confidence
7	GPT-6 Sol	71.9%	estimated ± 14.8 pp, low confidence
8	GPT-5.5 Pro	71.9%	estimated ± 14.8 pp, low confidence
9	GPT-5.4 Pro	71.9%	estimated ± 14.8 pp, low confidence
10	GPT-5.6 Terra	71.9%	estimated ± 14.8 pp, low confidence
11	Claude Opus 5	71.9%	estimated ± 14.8 pp, low confidence
12	Claude Fable 5	71.9%	estimated ± 14.8 pp, low confidence
13	Muse Wonder	71.9%	estimated ± 14.8 pp, low confidence
14	GPT-6.1 Terra	71.9%	estimated ± 14.8 pp, low confidence
15	GPT-5.5	71.9%	estimated ± 14.8 pp, low confidence
16	MiMo-V2.6-Pro	71.9%	estimated ± 14.8 pp, low confidence
17	Gemini 3 Pro Deep Think	71.9%	estimated ± 14.8 pp, low confidence
18	Muse Spark 1.3	71.9%	estimated ± 14.8 pp, low confidence
19	Kimi K3.1 Pro	71.9%	estimated ± 14.8 pp, low confidence
20	GPT-5.4	71.9%	estimated ± 14.8 pp, low confidence
21	Kimi K3	71.9%	estimated ± 14.8 pp, low confidence
22	MiniMax M3.1 Flash	71.9%	estimated ± 14.8 pp, low confidence
23	Claude Opus 4.8	71.9%	estimated ± 14.8 pp, low confidence
24	GLM-5.2	71.9%	estimated ± 14.8 pp, low confidence
25	Step 5 Preview	71.9%	estimated ± 14.8 pp, low confidence
26	GPT-5.6 Luna	71.9%	estimated ± 14.8 pp, low confidence
27	Sage-A1.5	71.9%	estimated ± 14.8 pp, low confidence
28	GPT-6 Luna	71.9%	estimated ± 14.8 pp, low confidence
29	GLM-5.3	71.9%	estimated ± 14.8 pp, low confidence
30	Claude Haiku 5.5	71.9%	estimated ± 14.8 pp, low confidence
31	Gemini 3.8 Flash	71.9%	estimated ± 14.8 pp, low confidence
32	GPT-6 Astra	71.8%	measured
33	Claude Opus 5.5	66.8%	measured
34	Claude Fable 5.1	65.0%	measured

GraphWalks BFS 256K–1M leaderboard

All leaderboards

Agentic · terminal

Agentic · tools

Coding

Math

Knowledge & reasoning

Instruction following

Multilingual

Vision & documents

Long context

General