OpenAI MRCR v2 8-needle 128K-256K (MRCR v2 128K-256K)
We show this table for reference; we do not rank on it.
MRCR v2 slice focused on very long contexts at 128K-256K lengths.
Benchmark score on MRCR v2 128K-256K — October 29, 2026
We compile the MRCR v2 128K-256K rows from provider self-reports and secondary reports. GPT-5.5 leads the table at 87.5%, followed by Claude Opus 4.7 (Adaptive) (59.2%). We do not use these results to rank models overall.
GPT-5.5
OpenAI
Claude Opus 4.7 (Adaptive)
Anthropic
2 modelsReasoningCurrentDisplay onlyUpdated October 29, 2026
Benchmark score table (2 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-5.5OpenAI | 87.5% | Not reported | Closed |
| 2 | Claude Opus 4.7 (Adaptive)Anthropic | 59.2% | Not reported | Closed |
About MRCR v2 128K-256K
Year
2026
Tasks
8-needle retrieval tasks
Format
Very-long-context retrieval
Difficulty
Very long-context reasoning
A harder MRCR setting that stresses memory discipline and retrieval deeper into long contexts.
Freshness and provenance
Version
MRCR v2 128K-256K 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does MRCR v2 128K-256K measure?
MRCR v2 slice focused on very long contexts at 128K-256K lengths.
Which model scores highest on MRCR v2 128K-256K?
GPT-5.5 by OpenAI currently leads with a score of 87.5% on MRCR v2 128K-256K.
How many models are evaluated on MRCR v2 128K-256K?
2 AI models have published results on MRCR v2 128K-256K in the BenchLM catalog.
Compare top models on MRCR v2 128K-256K
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.