benchmark
.
space
benchmarks
rankings
compare
voices
transcripts
papers
articles
MRCR v2 leaderboard
MRCR v2
18 models tested · Updated 2025-12-11 · Verified sources only
GPT-5.2
leads at
98.0%
1
GPT-5.2
OpenAI ·
Vellum — GPT-5.2 Benchmarks
· 2025-12-11
4-needle variant at 128k context. 8-needle 128k: 85%.
98.0%
2
Claude Opus 4.6
Anthropic ·
Anthropic — Introducing Claude Opus 4.6
· 2026-02-05
Highest published match ratio at 256k context. Drops to 76% at 1M tokens.
93.0%
3
Gemini 3.1 Pro
Google ·
Blog/Google DeepMind
· 2026-02-19
Long-context retrieval. Strong at 128k, drops to 26.3% at 1M tokens.
84.9%
4
DeepSeek V4 Pro
DeepSeek ·
DeepSeek/HuggingFace
· 2026-04-24
1M context. Below Opus 4.6 (92.9) but competitive.
83.5%
5
DeepSeek V4 Flash
DeepSeek ·
HuggingFace/deepseek-ai
· 2026-04-24
Solid long-context retrieval at 1M tokens for 13B activated.
78.7%
6
Gemini 3.5 Flash
Google DeepMind ·
Blog/Google DeepMind
· 2026-06-10
128k 8-needle average. Below Gemini 3.1 Pro (84.9).
77.3%
7
Hunyuan Hy3
Tencent ·
HuggingFace/tencent-Hy3-modelcard
· 2026-07-06
Up from 42.9% in Hy3 Preview. Major long-context improvement.
75.1%
8
GPT-5.5
OpenAI ·
OpenAI Blog
· 2026-04-23
512K-1M range. Massive lead over GPT-5.4 (36.6) and Opus 4.6 (32.2).
74.0%
9
Gemini 3 Flash
Google DeepMind ·
Blog/Google DeepMind
· 2026-06-10
128k 8-needle. Below Gemini 3.5 Flash (77.3).
67.2%
10
Gemma 4 31B
Google ·
Model Card/Google
· 2026-04-02
8-needle variant at 128k context. Solid long-context retrieval for a 31B dense model.
66.4%
11
Claude Opus 4.7
Anthropic ·
Blog/Google DeepMind
· 2026-06-10
128k 8-needle. Below Gemini 3.1 Pro (84.9).
59.3%
12
Claude Opus 4.7
Anthropic ·
Blog/OpenAI
· 2026-04-16
OpenAI-tested at 128K-256K. Much weaker long-context than GPT-5.5.
59.2%
13
MiMo-V2-Flash
Xiaomi ·
HuggingFace/XiaomiMiMo-MiMo-V2-Flash
· 2026-01-06
Long-context retrieval. Lower than frontier models (89.7).
45.7%
14
Gemma 4 26B A4B
Google ·
HuggingFace/Google DeepMind
· 2026-04-02
Multi-round co-reference resolution, 128k context.
44.1%
15
GPT-5.4
OpenAI ·
OpenAI Blog
· 2026-04-23
512K-1M range. Massive gap to GPT-5.5 (74.0).
36.6%
16
GLM-5
Zhipu ·
DocsBot AI benchmark comparison
· 2026-02-11
Sharp degradation at extreme context length. Drops from 77% at 128k.
26.3%
17
Gemma 4 E4B
Google ·
HuggingFace/Google DeepMind
· 2026-04-02
Multi-round co-reference resolution, 128k context.
25.4%
18
Gemma 4 E2B
Google ·
HuggingFace/Google DeepMind
· 2026-04-02
Multi-round co-reference resolution, 128k context.
19.1%