benchmarks.site

Public leaderboard · Long context

Context Arena (MRCR)

Hides several near-identical items in a very long conversation and asks the AI to retrieve a specific one, checking whether it can find the right detail in a huge pile of text.

Leaderboard

Top 3 right now

  1. 1stopenai/gpt-5.6-sol [max]92.4
  2. 2ndqwen/qwen3.8-max [xhigh]92.3
  3. 3rdanthropic/claude-opus-5 [max]91.3

What it measures

Leaderboard for OpenAI's MRCR test, where the model must reproduce a specific one of several similar requests buried in a long conversation, scored at multiple context lengths.

How to read the numbers

Average retrieval score (%) on the 8-needle task for conversations of 64k-128k tokens — higher is better.

Results

25 rows · higher is better (Score at 64k-128k tokens (%))CSV
1openai/gpt-5.6-sol [max]92.485.5
2qwen/qwen3.8-max [xhigh]92.373.7
3anthropic/claude-opus-5 [max]91.377.3
4google/gemini-3.7-flash [high]88.683.0
5qwen/qwen3.8-max [medium]87.372.1
6openai/gpt-5.5 [xhigh]85.879.8
7openai/gpt-5.5 [medium]83.078.6
8google/gemini-3.7-flash [medium]82.179.6
9qwen/qwen3.8-27b [low]81.892.8
10qwen/qwen3.8-max [low]81.571.6
11openai/gpt-5.6-terra [max]80.276.6
12openai/gpt-5.5 [high]79.379.0
13qwen/qwen3.8-27b [medium]78.994.0
14google/gemini-3.7-flash [low]78.374.5
15google/gemini-3.6-flash [high]77.373.3
16anthropic/claude-opus-4.8 [max]75.272.6
17google/gemini-3.6-flash [medium]74.569.9
18qwen/qwen3.8-27b [xhigh]74.390.2
19anthropic/claude-sonnet-4.6 [low]72.670.4
20anthropic/claude-opus-4.6 [medium]72.573.1
21anthropic/claude-sonnet-4.6 [high]72.470.5
22anthropic/claude-sonnet-4.6 [medium]71.769.6
23qwen/qwen3.6-max-preview [enabled]71.688.1
24z-ai/glm-5.3 [high]69.370.4
25anthropic/claude-opus-4.6 [low]68.872.4

Note: 8-needle, full mode, from the site API (184 entries). Ranked here by the 64k-128k token bin because every model has it; the site API gives no ordering and no update date. Overall score averages over whichever lengths each model was tested at, so is not directly comparable. The 2-needle view returned empty at retrieval.

Source: Context Arena · contextarena.ai · retrieved 2026-11-17

Markdown · More Long context