Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

See the free Radar Brief

OpenAI MRCR v2 8-needle 256K-512K (MRCR v2 256K-512K)

MRCR v2 slice focused on retrieval across 256K-512K-token contexts.

Data verified 27 confirmed releases in the last 30 daysSee the free Radar Brief

Benchmark score on MRCR v2 256K-512K — September 2, 2026

We mirror the published score view for MRCR v2 256K-512K. Muse Spark 1.3 leads the public snapshot at 98.5%. We do not use these results to rank models overall.

1 modelReasoningCurrentDisplay onlyUpdated September 2, 2026

Benchmark score table (1 model)

Score
1
Muse Spark 1.3Meta · Closed
98.5%

About MRCR v2 256K-512K

Year

2026

Tasks

100 eight-needle retrieval examples

Format

Mean sequence-matcher ratio

Difficulty

Very long-context retrieval

Meta re-bins OpenAI's MRCR v2 data by o200k_base token count and evaluates 100 examples with eight needles. The mean sequence-matcher ratio stays display-only because this slice is not part of the weighted reasoning schema.

BenchLM freshness & provenance

Version

MRCR v2 256K-512K 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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.

FAQ

What does MRCR v2 256K-512K measure?

MRCR v2 slice focused on retrieval across 256K-512K-token contexts.

Which model scores highest on MRCR v2 256K-512K?

Muse Spark 1.3 by Meta currently leads with a score of 98.5% on MRCR v2 256K-512K.

How many models are evaluated on MRCR v2 256K-512K?

1 AI models have been evaluated on MRCR v2 256K-512K on BenchLM.

Last updated: September 2, 2026 · BenchLM version MRCR v2 256K-512K 2026

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.