Chronicles-OCR - Mature Text Parsing: leaderboard

Chronicles-OCR (2026) tests vision-language models on historical Chinese writing across the Seven Chinese Scripts, 2,800 expert-annotated images from Oracle Bone to Cursive. This board scores transcription of Clerical, Regular, Running and Cursive texts in their original reading order, as normalized edit similarity from 0 to 1 averaged over the four scripts; higher is better.

Metric: Paragraph-level normalized edit similarity (0-1: one minus the edit distance over the longer sequence) of the transcription in reading order against the reference, on the mature scripts (Clerical, Regular, Running and Cursive), the paper Average column over the per-script results; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 26 models tracked. Updated 7 Oct 2026.

Top models

#ModelSettingScoreOverall rank
1Qwen 3.5 397B A17BNon-reasoning0.73#264
2Seed 2.0 ProNon-reasoning0.72#222
3Kimi K2.5Non-reasoning0.71#266
4Qwen 3.5 35B A3BNon-reasoning0.71#492
5Gemini 3.1 Pro (Preview)0.7#99
6Seed 1.8Non-reasoning0.69#252
7Qwen 3 VL 8B Instruct0.66#1029
8Qwen 3 VL 235B A22B Instruct0.66#532
9Qwen 3 VL 235B A22B (Thinking)0.65#459
10Qwen 3 VL 2B Instruct0.57#2367
11InternVL3.5-241B-A28B0.56#733
12MiMo-V2-Omni0.56#465
13Gemini 2.5 Pro0.53#276
14Ovis2.6-30B-A3BThinking0.53
15Claude Opus 4.7Thinking0.5#85

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=chronicles-ocr-mature-text-parsing · How It Works · Data refreshed daily, snapshot 2026-11-21.