ResearchLLM Verifications Leaderboard

Data as of · 78 models, 80 questions · reviewed by Suhas Bhairav · methodology

Best LLM for…

The best LLMs with a 1M-token context window or more

Answer: Claude Opus 5.5 by Anthropic: 85.0% with a 1M-token context window.

Reviewed by Suhas Bhairav · Data as of

Ranking

Models whose context window is at least 1,000,000 tokens (from OpenRouter's catalogue), ordered by verification score. The leaderboard doesn't test long-context recall itself. Top 15 of 37.

The best LLMs with a 1M-token context window or more
#ModelScoreContextCost / questionSpeed (tok/s)
1Claude Opus 5.5Anthropic85.0%1M$0.015129
2Gemini 4 ArgonGoogle83.8%1M$0.006269
3Gemini 3.8 FlashGoogle82.5%1M$0.0037142
4Claude Fable 5.1Anthropic81.2%1M$0.01670
5Muse WonderMeta78.8%1M$0.021175
6Claude Sonnet 5.5Anthropic78.8%1M$0.0054126
7Gemini 3.1 Pro PreviewGoogle77.5%1M$0.017124
8Muse Spark OpenMeta77.5%1M$0.0045220
9Kimi K3.1 ProMoonshotAI77.5%1M$0.0138
10GPT-6 AstraOpenAI76.2%1.1M$0.01349
11GPT-6.1 Terra ProOpenAI76.2%1.1M$0.005—
12GPT-6.1 SolOpenAI76.2%1.1M$0.002941
13GPT-5.6 TerraOpenAI75.0%1.1M$0.0063101
14Gemini 3.5 Flash LiteGoogle75.0%1M$0.0034409
15Qwen3.8 Max (0902)Qwen75.0%1M$0.01146

How to read this

Scores come from 80 multiple-choice questions from SuperGPQA and MMLU-Pro. With that many questions, a score has a margin of error of several points, so models a few points apart are effectively tied on accuracy; the comparison pages run a significance test for any pair. To see the top two side by side, compare Claude Opus 5.5 and Gemini 4 Argon.

Speed, latency and cost are measured through OpenRouter and depend on which host serves the request; your numbers on a direct API may differ. The full procedure is on the methodology page, and the data behind this ranking can be downloaded.

Other questions

Quick answers

What is the most accurate AI model with a million-token context window?
Claude Opus 5.5 by Anthropic: 85.0% with a 1M-token context window. Rule: Models whose context window is at least 1,000,000 tokens (from OpenRouter's catalogue), ordered by verification score. The leaderboard doesn't test long-context recall itself.
What is the runner-up?
Gemini 4 Argon by Google: 83.8% with a 1M-token context window.
Which of these costs least per correct answer?
GLM 5.3 Flash: $0.00022 per correct answer (cost per question divided by its score).