Spark-X2.6 Release - Openai_MRCR v2 256k-512k: leaderboard
Spark-X 2.6 official release SVG; own unstarred evaluations only, source-starred reprints and missing cells not duplicated. All runs thinking; Spark-X 2.6 explicitly high, recommended temperature 1/top_p 0.95/top_k-1. Other models remain thinking with effort unspecified. Gemma 4-12B label is source-scoped because exact checkpoint is not identified; never aliased to Gemma 3/4 or inferred weights. Gaokao 2026 averages five Chinese exams, each 150 points; other source benchmark scores retain native 0–100 units. No undocumented decoding/harness assumed. Metric: Score (%)
Metric: Score (%). Source: huggingface.co. Saturation forecast: Around February 2027. 4 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 9B (Thinking) | 58.6 | #391 (Qwen 3.5 9B) |
| 2 | Qwen 3.5 4B (Thinking) | 48.3 | #514 (Qwen 3.5 4B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=spark-x2-6-release-openai-mrcr-v2-256k-512k · How It Works · Data refreshed daily, snapshot 2026-11-05.