BlogLLM Benchmark Scores as Published, Side by Side
MRCR: Results as Published
Every result the catalog holds for MRCR: 36 models from 6 developers, read from 14 pages. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area Long ContextMaintainer Kiran Vodrahalli and co-authors of the Michelangelo paperIntroduced 2024Data version 2026103001, as of 2026-10-30
MRCR (Multi-Round Coreference Resolution) tests whether a model can find one specific earlier item in a very long chat. The conversation contains many similar requests, such as poems or essays on overlapping topics, and the model must reproduce the output that matches the one the user asked for. The 8-needle variant hides eight such matching items, and results are reported at long context lengths such as 256K, 512K and 1M tokens.
MRCR v2 8-needle 128K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.6 LunaOpenAI, released 2026-07-09 | 74.8% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15 | Context: 128k (average) |
| GPT-5.6 TerraOpenAI, released 2026-07-09 | 93.5% | Google, Gemini 3.7 Flash model cardchecked 2026-10-15 | - |
| GPT-5.5OpenAI, released 2026-04-23 | 94.8% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | Context: 128k (average) |
| GPT-5.4 miniOpenAI, released 2026-03-17 | 42.7% | Google, Gemini 3.5 Flash-Lite model cardchecked 2026-10-15 | Context: 128k (average) |
| GPT-5.2OpenAI, released 2025-12-11 | 83.8% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15 | Effort: xhigh; Context: 128k (average) |
| Claude Sonnet 5Anthropic, released 2026-06-30 | 71.6% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15; printed on 1 more page | Context: 128k (average) |
| Claude Opus 4.7Anthropic, released 2026-04-16 | 59.3% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | Context: 128k (average) |
| Claude Sonnet 4.6Anthropic, released 2026-02-17 | 84.9% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 1 more page | Effort: max; Context: 128k (average) |
| Claude Opus 4.6Anthropic, released 2026-02-05 | 84% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15 | Effort: max; Context: 128k (average) |
| Claude Haiku 4.5Anthropic, released 2025-10-15 | 35.3% | Google, Gemini 3.5 Flash-Lite model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemini 3.7 FlashGoogle, released 2026-08-13 | 97% | Google, Gemini 3.7 Flash model cardchecked 2026-10-15 | - |
| Gemini 3.5 Flash-LiteGoogle, released 2026-07-21 | 72.2% | Google, Gemini 3.5 Flash-Lite model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemini 3.6 FlashGoogle, released 2026-07-21 | 91.8% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15; printed on 1 more page | Context: 128k (average) |
| Gemini 3.5 FlashGoogle, released 2026-05-19 | 77.3% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15; printed on 1 more page | Context: 128k (average) |
| Gemma 4 12BGoogle, released 2026-04-02 | 43.4% | Google, Gemma 4 model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemma 4 26B A4BGoogle, released 2026-04-02 | 44.1% | Google, Gemma 4 model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemma 4 31BGoogle, released 2026-04-02 | 66.4% | Google, Gemma 4 model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemma 4 E2BGoogle, released 2026-04-02 | 19.1% | Google, Gemma 4 model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemma 4 E4BGoogle, released 2026-04-02 | 25.4% | Google, Gemma 4 model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemini 3.1 Flash-LiteGoogle, released 2026-03-03 | 60.1% | Google, Gemini 3.5 Flash-Lite model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemini 3.1 Pro PreviewGoogle, released 2026-02-19 | 84.9% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 2 more pages | Effort: high; Context: 128k (average) |
| Gemini 3 FlashGoogle, released 2025-12-17 | 67.2% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | Context: 128k (average) |
| Gemini 3 ProGoogle, released 2025-11-18 | 77% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15 | Effort: high; Context: 128k (average) |
| Gemma 3 27BGoogle, released 2025-03-12 | 13.5% | Google, Gemma 4 model cardchecked 2026-10-15 | Effort: no think; Context: 128k (average) |
| Grok 4.5xAI, released 2026-07-16 | 81.4% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15 | Context: 128k (average) |
MRCR v2 8-needle 128K-256K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.5OpenAI, released 2026-04-23 | 87.5% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 128K-256K |
| GPT-5.4 miniOpenAI, released 2026-03-17 | 33.6% | OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15 | Effort: xhigh; Context: 128K-256K |
| GPT-5.4 nanoOpenAI, released 2026-03-17 | 33.1% | OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15 | Effort: xhigh; Context: 128K-256K |
| GPT-5.4OpenAI, released 2026-03-05 | 79.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 2 more pages | Effort: xhigh; Context: 128K-256K |
| GPT-5.2OpenAI, released 2025-12-11 | 77% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 128k-256k |
| GPT-5.1OpenAI, released 2025-11-12 | 29.6% | OpenAI, Introducing GPT-5.2checked 2026-10-15 | Effort: high; Context: 128k-256k |
| GPT-5 miniOpenAI, released 2025-08-07 | 19.4% | OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15 | Effort: high; Context: 128K-256K |
| Claude Opus 4.7Anthropic, released 2026-04-16 | 59.2% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Context: 128K-256K |
MRCR v2 8-needle 1M (pointwise)
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Gemini 3.5 Flash-LiteGoogle, released 2026-07-21 | 21.3% | Google, Gemini 3.5 Flash-Lite model cardchecked 2026-10-15 | Context: 1M (pointwise) |
| Gemini 3.6 FlashGoogle, released 2026-07-21 | 54% | Google, Gemini 3.6 Flash model cardchecked 2026-10-15 | Context: 1M (pointwise) |
| Gemini 3.5 FlashGoogle, released 2026-05-19 | 26.6% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15; printed on 1 more page | Context: 1M (pointwise) |
| Gemini 3.1 Flash-LiteGoogle, released 2026-03-03 | 12.3% | Google, Gemini 3.5 Flash-Lite model cardchecked 2026-10-15 | Context: 1M (pointwise) |
| Gemini 3.1 Pro PreviewGoogle, released 2026-02-19 | 26.3% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15; printed on 2 more pages | Effort: high; Context: 1M (pointwise) |
| Gemini 3 FlashGoogle, released 2025-12-17 | 22.1% | Google, Gemini 3.5 Flash model cardchecked 2026-10-15 | Context: 1M (pointwise) |
| Gemini 3 ProGoogle, released 2025-11-18 | 26.3% | Google, Gemini 3.1 Pro model cardchecked 2026-10-15 | Effort: high; Context: 1M (pointwise) |
MRCR v2 8-needle 512K-1M
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6 AstraOpenAI, released 2026-09-03 | 96.3% | OpenAI, GPT-6 Astrachecked 2026-10-14 | - |
| GPT-5.6 LunaOpenAI, released 2026-07-09 | 41.3% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15 | Context: 512K-1M |
| GPT-5.6 SolOpenAI, released 2026-07-09 | 73.8% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15; printed on 1 more page | Context: 512K-1M |
| GPT-5.6 TerraOpenAI, released 2026-07-09 | 72.5% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15 | Context: 512K-1M |
| GPT-5.5OpenAI, released 2026-04-23 | 74% | OpenAI, Introducing GPT-5.5checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 512K-1M |
| GPT-5.4OpenAI, released 2026-03-05 | 36.6% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 512K-1M |
| Claude Opus 4.7Anthropic, released 2026-04-16 | 32.2% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Context: 512K-1M |
MRCR v2 8-needle 64K-128K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.5OpenAI, released 2026-04-23 | 83.1% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 64K-128K |
| GPT-5.4 miniOpenAI, released 2026-03-17 | 47.7% | OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15 | Effort: xhigh; Context: 64K-128K |
| GPT-5.4 nanoOpenAI, released 2026-03-17 | 44.2% | OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15 | Effort: xhigh; Context: 64K-128K |
| GPT-5.4OpenAI, released 2026-03-05 | 86% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 2 more pages | Effort: xhigh; Context: 64K-128K |
| GPT-5.2OpenAI, released 2025-12-11 | 85.6% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 64k-128k |
| GPT-5.1OpenAI, released 2025-11-12 | 36% | OpenAI, Introducing GPT-5.2checked 2026-10-15 | Effort: high; Context: 64k-128k |
| GPT-5 miniOpenAI, released 2025-08-07 | 35.1% | OpenAI, Introducing GPT-5.4 mini and nanochecked 2026-10-15 | Effort: high; Context: 64K-128K |
MRCR v2 8-needle 256K-512K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-6 AstraOpenAI, released 2026-09-03 | 100% | OpenAI, GPT-6 Astrachecked 2026-10-14 | - |
| GPT-5.6 LunaOpenAI, released 2026-07-09 | 41.3% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15 | Context: 256K-512K |
| GPT-5.6 SolOpenAI, released 2026-07-09 | 91.5% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15; printed on 1 more page | Context: 256K-512K |
| GPT-5.6 TerraOpenAI, released 2026-07-09 | 89.6% | OpenAI, GPT-5.6: Frontier intelligence that scales with your ambitionchecked 2026-10-15 | Context: 256K-512K |
| GPT-5.5OpenAI, released 2026-04-23 | 81.5% | OpenAI, Introducing GPT-5.5checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 256K-512K |
| GPT-5.4OpenAI, released 2026-03-05 | 57.5% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 256K-512K |
MRCR v2 8-needle 16K-32K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.5OpenAI, released 2026-04-23 | 96.5% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 16K-32K |
| GPT-5.4OpenAI, released 2026-03-05 | 97.2% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 16K-32K |
| GPT-5.2OpenAI, released 2025-12-11 | 95.3% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 16k-32k |
| GPT-5.1OpenAI, released 2025-11-12 | 44% | OpenAI, Introducing GPT-5.2checked 2026-10-15 | Effort: high; Context: 16k-32k |
MRCR v2 8-needle 32K-64K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.5OpenAI, released 2026-04-23 | 90% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 32K-64K |
| GPT-5.4OpenAI, released 2026-03-05 | 90.5% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 32K-64K |
| GPT-5.2OpenAI, released 2025-12-11 | 92% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 32k-64k |
| GPT-5.1OpenAI, released 2025-11-12 | 37.8% | OpenAI, Introducing GPT-5.2checked 2026-10-15 | Effort: high; Context: 32k-64k |
MRCR v2 8-needle 4K-8K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.5OpenAI, released 2026-04-23 | 98.1% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 4K-8K |
| GPT-5.4OpenAI, released 2026-03-05 | 97.3% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 4K-8K |
| GPT-5.2OpenAI, released 2025-12-11 | 98.2% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 4k-8k |
| GPT-5.1OpenAI, released 2025-11-12 | 65.3% | OpenAI, Introducing GPT-5.2checked 2026-10-15 | Effort: high; Context: 4k-8k |
MRCR v2 8-needle 8K-16K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.5OpenAI, released 2026-04-23 | 93% | OpenAI, Introducing GPT-5.5checked 2026-10-15 | Effort: xhigh; Context: 8K-16K |
| GPT-5.4OpenAI, released 2026-03-05 | 91.4% | OpenAI, Introducing GPT-5.4checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 8K-16K |
| GPT-5.2OpenAI, released 2025-12-11 | 89.3% | OpenAI, Introducing GPT-5.2checked 2026-10-15; printed on 1 more page | Effort: xhigh; Context: 8k-16k |
| GPT-5.1OpenAI, released 2025-11-12 | 47.8% | OpenAI, Introducing GPT-5.2checked 2026-10-15 | Effort: high; Context: 8k-16k |
MRCR 1M (MMR)
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| Claude Opus 4.6Anthropic, released 2026-02-05 | 92.9% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max; Context: 1M |
| Gemini 3.1 Pro PreviewGoogle, released 2026-02-19 | 76.3% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high; Context: 1M |
| DeepSeek-V4-Flash (Preview)DeepSeek, released 2026-04-24 | 76.9% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high; Context: 1M |
| 78.7% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max; Context: 1M | |
| 37.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: non-think; Context: 1M | |
| DeepSeek-V4-Pro (Preview)DeepSeek, released 2026-04-24 | 44.7% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: non-think; Context: 1M |
| 83.5% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: max; Context: 1M | |
| 83.3% | DeepSeek, DeepSeek-V4-Pro (preview) model cardchecked 2026-10-15 | Effort: high; Context: 1M |
MRCR v2 8-needle 256K
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5.6 SolOpenAI, released 2026-07-09 | 93.8 | Alibaba, Qwen3.8-Max comparisonchecked 2026-10-14 | Effort: max |
| Claude Opus 4.8Anthropic, released 2026-05-28 | 83.2 | Alibaba, Qwen3.8-Max comparisonchecked 2026-10-14 | - |
| Qwen3.8-MaxAlibaba, released 2026-08-03 | 92.9 | Alibaba, Qwen3.8-Max comparisonchecked 2026-10-14 | - |
| Qwen3.7-MaxAlibaba | 86.7 | Alibaba, Qwen3.8-Max comparisonchecked 2026-10-14 | - |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.