Draw the chart from its title
Recall ten years of monthly Wikipedia pageviews from the article title alone.
Run Nov 10, 2026, 197 items.
Method
Each model sees only an English Wikipedia article title and draws its 120 monthly pageview values from January 2016 through December 2025. It receives no chart, page text, real values or tools. The same 197 articles are scored for every model; this tests recall of real-world quantities and events, not visual chart reading.
Articles come from a seeded sample of an October 2025 pageview dump, keeping complete ten-year series with a median of at least 3,000 monthly views. Eight articles renamed inside the window were removed so their missing earlier-title traffic would not count against a model.
One answer per chart at the provider's default reasoning setting, with one retry if it cannot be parsed. All 197 charts count: a final unparsable answer receives that article's flat-line reference score. OpenRouter calls allow 32,000 output tokens including reasoning. Claude models use the Claude CLI without a recorded output limit. Fable 5.2 and Carbon each reuse 28 answers to identical pilot prompts; Carbon's reused answers used a 16,000-token limit and finished below it.
The score is the mean, across the 197 articles, of the root mean squared difference between predicted and real log10 monthly views. Lower is better; 0 is exact. An error of 0.30 corresponds to a typical multiplicative miss of about 2×. This score measures both the article's popularity level and the curve's shape.
Results
| # | Model | RMSE (log10 views) (lower is better) |
|---|---|---|
| 1 | GPT-6.1 Terra (OpenRouter · 32,000-token output limit) | 0.272468 |
| 2 | Gemini 4 Carbon (OpenRouter · 32,000-token output limit (16,000 for 28 reused pilot answers)) | 0.272598 |
| 3 | Gemini 3.8 Flash (OpenRouter · 32,000-token output limit) | 0.279252 |
| 4 | Claude Opus 5.5 (Claude CLI · output limit not recorded) | 0.285835 |
| 5 | Claude Fable 5.2 (Claude CLI · output limit not recorded) | 0.291679 |
| Average article (baseline) | 0.402218 | |
| Flat line (baseline) | 0.403034 | |
| 6 | GLM-5.3 Flash (OpenRouter · 32,000-token output limit) | 0.410944 |
| 7 | Muse Spark 1.3 (OpenRouter · 32,000-token output limit) | 0.412021 |
| 8 | MiMo-V2.6-Pro (OpenRouter · 32,000-token output limit) | 0.425083 |
| 9 | DeepSeek V4.1 Flash (1017) (OpenRouter · 32,000-token output limit) | 0.532144 |
| 10 | Muse Ember (OpenRouter · 32,000-token output limit) | 0.533896 |
| 11 | Mistral Large 4 (OpenRouter · 32,000-token output limit) | 0.572447 |
| 12 | MiniMax-M3.1-Flash (OpenRouter · 32,000-token output limit) | 0.594623 |
| 13 | Claude Sonnet 5.5 (Claude CLI · output limit not recorded) | 0.674311 |
| 14 | Qwen 3.8 Flash (OpenRouter · 32,000-token output limit) | 0.730670 |
| 15 | GPT-6 Luna (OpenRouter · 32,000-token output limit) | 0.805774 |
| 16 | Claude Haiku 5.5 (Claude CLI · output limit not recorded) | 0.930764 |
- The reference predictors know nothing about the article. They use the geometric mean of the other 196 articles: their average monthly curve, or a constant level as a flat line. They are references, not model observations.
- The sample favors articles read around midday UTC in one October 2025 dump. Each model ran once. The study's chart-bootstrap intervals measure variation across articles, not variation from asking a model again.
Charts, exact prompt and every prediction
In the main leaderboard
These results are collected for the main leaderboard alongside external benchmarks. New scores enter the rating through the regular audited data updates. Comparisons with that rating therefore share some of its inputs.
Interactive version: theaggregate.ai/our-benchmarks/chart-recall · How It Works · Data refreshed daily, snapshot 2026-11-12.