MRCR v2 leaderboard
As of November 8, 2026, Gemini 3.7 Flash (Google DeepMind) leads MRCR v2 at 97%, ahead of Gemini 3.6 Flash at 91.8% and Gemini 3.1 Pro Preview at 84.9%. 7 systems ranked. Every score here is published by the model's own lab; each row links to the page it comes from.
What is MRCR v2? MRCR v2 (multi-round coreference resolution) tests whether a model can find one specific earlier reply in a very long conversation full of near-identical ones and reproduce it exactly. Made by Google DeepMind, February 2026 (task introduced in 2024).
Top models on MRCR v2, November 2026
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | Gemini 3.7 Flash Deprecated Google DeepMind | 128K average | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
| 2 | Gemini 3.6 Flash Google | 128K average | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
| 3 | Gemini 3.1 Pro Preview Preview Google · listed as “Gemini 3.1 Pro” | 128K average | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 4 | Gemini 3.5 Flash Deprecated Google | 128K average | $1.5 / $9 | 1M | Vendor-reported Google DeepMind |
| 5 | Gemini 3 Pro Preview Retired Google · listed as “Gemini 3 Pro” | 128K average, thinking high | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 6 | Gemini 3.5 Flash-Lite Google | 128K average | $0.3 / $2.5 | 1M | Vendor-reported Google DeepMind |
| 7 | Gemini 3.1 Flash-Lite Deprecated Google | 128K average | $0.25 / $1.5 | 1M | Vendor-reported Google DeepMind |
MRCR v2 under other settings
Labs report MRCR v2 under different settings, and a score from one setting is not comparable with a score from another. Each setting has its own table, ranked on its own.
8-needle, at 1M
| # | Model | Score | Price / 1M | Context | Source |
|---|---|---|---|---|---|
| 1 | Gemini 3.6 Flash Google | 1M pointwise | $0.75 / $3.75 | 1M | Vendor-reported Google DeepMind |
| 2 | Gemini 3.5 Flash Deprecated Google | 1M pointwise | $1.5 / $9 | 1M | Vendor-reported Google DeepMind |
| 3 | Gemini 3 Pro Preview Retired Google · listed as “Gemini 3 Pro” | 1M pointwise, thinking high | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 3 | Gemini 3.1 Pro Preview Preview Google · listed as “Gemini 3.1 Pro” | 1M pointwise | $2 / $12 | 1M | Vendor-reported Google DeepMind |
| 5 | Gemini 3.5 Flash-Lite Google | 1M pointwise | $0.3 / $2.5 | 1M | Vendor-reported Google DeepMind |
| 6 | Gemini 3.1 Flash-Lite Deprecated Google | 1M pointwise | $0.25 / $1.5 | 1M | Vendor-reported Google DeepMind |
Can you trust MRCR v2 scores?
No independent leaderboard publishes MRCR v2 results for these models yet, so every number on this page is the lab's own report of its own model. Labs choose the harness, the effort level and the subset they report, and they rarely publish failures. Read the setting under each score before comparing two rows.
- Google reports the 8-needle version at two context lengths that are defined differently (cumulative up to 128K, and at 1M), so they are separate tables here.
- With code tools the task gets much easier; the authors ask that every report says whether tools were used.
- Random guessing among the similar replies already earns a non-trivial score, so low numbers sit near a floor.
How MRCR v2 works
The task
The model reads a long synthetic conversation where the user asks many times for writing in a given format and topic; it has to reproduce one particular earlier reply, among two, four or eight similar ones.
The score
A string-similarity score between the reproduced text and the target, set to zero if the required random prefix is missing.
Test set
The authors do not publish a single task count for this version.
Made by
Google DeepMind, first published February 2026 (task introduced in 2024).
Official sources
- github.com/google-deepmind/eval_hub/tree/master/eval_hub/mrcr_v2
- Paper: arxiv.org/abs/2409.12640
- Definition read on 2026-11-08
Sources and freshness
13 published scores from 4 sources. Each was read on its source page; the oldest reading on this page is from November 8, 2026. Prices beside each score come from our model price index and move when a vendor reprices.
- Vendor-reported Google DeepMind
- Vendor-reported Google DeepMind
- Vendor-reported Google DeepMind
- Vendor-reported Google DeepMind
MRCR v2 FAQ
Answers derived from the 13 published scores on this page and the current price index.