Skip to content

Benchmark

OpenAI MRCR v2 · 8-needle

Long context benchmarks measure how well models use information in long inputs. Scores are percentages. Higher is better.

Models measured
28
Scores recorded
74
Sources
1
IQ weight
1.76

Score distribution

78.1%
Mean
84.9%
Median
19.3%
Std dev
44.1%
10th percentile
97.2%
90th percentile
26.3%
Min
26.3%100%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.76 for chat models, measured over 28 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
51GPT 5.5512K-1M· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.74%
52GPT 5.5512K–1M · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.74%
53GPT-5.6 Sol512K-1M· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.8%
54GPT-5.6 Solmax · 512K–1M · 100 examples · 8-needle · no tools · mean sequence-matcher ratio· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.73.8%
55GPT-5.6 Terra512K-1M· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.72.5%
56Claude Sonnet 5Gemini 3.6 Flash Model Card · July 2026 · 128K cumulative average· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.71.6%
57Gemini 3 Flash128K cumulative average · 8-needle· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.67.2%
58Gemini 3 FlashGemini 3.5 Flash Model Card · May 2026 · 128K cumulative average· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.67.2%
59Gemma 4 31B ITGemma 4 Technical Report · Table 5 · thinking mode unless noted · 128K· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66.4%
60Muse Spark 1.2xhigh · 256K–512K · 100 examples · 8-needle · no tools · mean sequence-matcher ratio· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.66.3%
61Gemini 3.1 Flash LiteGemini 3.1 Flash-Lite Model Card · high thinking · default sampling · single attempt/pass@1 · 128K cumulative average· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.60.1%
62Claude Opus 4.7Gemini 3.5 Flash Model Card · May 2026 · 128K cumulative average· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.59.3%
63GPT 5.4256K-512K· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.57.5%
64Muse Spark 1.2xhigh · 512K–1M · 100 examples · 8-needle · no tools · mean sequence-matcher ratio· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.55.5%
65GPT-5.4 minixhigh · 8-needle · 64K–128K· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.47.7%
66GPT 5.4 Nanoxhigh · 8-needle · 64K–128K· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.44.2%
67Gemma 4 26B A4BGemma 4 Technical Report · Table 5 · thinking mode unless noted · 128K· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.44.1%
68GPT-5.6 Luna256K-512K· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.41.3%
69GPT-5.6 Luna512K-1M· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.41.3%
70GPT 5.4512K-1M· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.36.6%
71GPT-5.4 minixhigh · 8-needle · 128K–256K· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.33.6%
72GPT 5.4 Nanoxhigh · 8-needle · 128K–256K· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.33.1%
73Step 3.5 FlashStep 3.5 technical report · Table 5 · 8-needle · vanilla· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.28.8%
74Step 3.5 FlashStep 3.5 technical report · Table 5 · 8-needle · PaCoRe· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.26.3%

Showing 51–74 of 74 results.

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.