QuantSightBench
Models give 90% prediction intervals for 1,000 numerical forecasting questions with known outcomes. The main setting allows retrieval from a fixed news corpus. Mean Log Interval Score penalizes wide intervals and missed outcomes, while coverage measures how often intervals contain the answer.
Questions derive from OpenForecast but require numerical intervals and a different scoring rule. This is a retrospective evaluation with restricted evidence dates, not a live forecast track record. Coverage has a 90% target and is not monotonically better when higher.
- Made by
- ELLIS Institute Tübingen, Max Planck Institute for Intelligent Systems, and Tübingen AI Center
- Released
- April 2026
- Kind
- Real-world events
- Level
- Expert
- Scored
- Settled by events
- Answers
- Prediction
- Tasks
- All public
- Measures
- Capability
- Subject
- Mathematics
- Language
- English
- Interaction
- Agentic
- Environment
- APIs and tools
- Items from
- Another benchmark
- Refresh
- One fixed set
- Run by
- Anyone
- Based on
- A version of another benchmark
- In views
- Agents
- Strongest model tested
- GPT-5.4 #28 on the capability index
- Last model added
- March 2026
- Scores refreshed
- November 19th, 2026
In the authors’ words
Forecasting has become a natural benchmark for reasoning under uncertainty. Yet existing evaluations of large language models remain limited to judgmental tasks in simple formats, such as binary or multiple-choice questions. In practice, however, forecasting spans a far broader scope. Across domains such as economics, public health, and social demographics, decisions hinge on numerical estimates over continuous quantities, a capability that current benchmarks do not capture. Evaluating such estimates requires a format that makes uncertainty explicit and testable. We propose prediction intervals as a natural and rigorous interface for this purpose. They demand scale awareness, internal consistency across confidence levels, and calibration over a continuum of outcomes, making them a more suitable evaluation format than point estimates for numerical forecasting. To assess this capability, we introduce a new benchmark QuantSightBench, and evaluate frontier models under multiple settings, assessing both empirical coverage and interval sharpness. Our results show that none of the 11 evaluated frontier and open-weight models achieves the 90% coverage target, with the top performers Gemini 3.1 Pro (79.1%), Grok 4 (76.4%), and GPT-5.4 (75.3%) all falling at least 10 percentage points short. Calibration degrades sharply at extreme magnitudes, revealing systematic overconfidence across all evaluated models.
GPT-5.1 leads with 6.70 among models released by March 2026
Best score per model from the benchmark's own leaderboard, read on November 19th, 2026. Where a model ran with several settings or harnesses, we keep its best.
The newest model with a score here came out in March 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 1 GPT-5.1 agentic, high reasoning effort 6.70
- 2 Kimi K2 Thinking agentic, high reasoning effort 6.81
- 3 Claude Opus 4.6 agentic, high reasoning effort 7.00
- 4 Claude Opus 4.5 agentic, high reasoning effort 7.07
- 5
Gemini 3.1 Pro agentic, high reasoning effort 7.26
- 6 GPT-5.4 agentic, high reasoning effort 7.32
- 7 Grok 4 agentic, high reasoning effort 7.44
- 8 GLM-4.7 agentic, high reasoning effort 7.49
- 9
Gemini 3 Pro agentic, high reasoning effort 7.52
- 10 Claude Sonnet 4.5 agentic, high reasoning effort 7.97
- 11
DeepSeek-V3.2 agentic, high reasoning effort 8.58
Source: Official QuantSightBench leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5.1 | 6.70 | agentic, high reasoning effort | Nov 2025 | Official leaderboard |
| 2 | Kimi K2 Thinking | 6.81 | agentic, high reasoning effort | Nov 2025 | Official leaderboard |
| 3 | Claude Opus 4.6 | 7.00 | agentic, high reasoning effort | Feb 2026 | Official leaderboard |
| 4 | Claude Opus 4.5 | 7.07 | agentic, high reasoning effort | Nov 2025 | Official leaderboard |
| 5 | | 7.26 | agentic, high reasoning effort | Feb 2026 | Official leaderboard |
| 6 | GPT-5.4 | 7.32 | agentic, high reasoning effort | Mar 2026 | Official leaderboard |
| 7 | Grok 4 | 7.44 | agentic, high reasoning effort | Jul 2025 | Official leaderboard |
| 8 | GLM-4.7 | 7.49 | agentic, high reasoning effort | Dec 2025 | Official leaderboard |
| 9 | | 7.52 | agentic, high reasoning effort | Nov 2025 | Official leaderboard |
| 10 | Claude Sonnet 4.5 | 7.97 | agentic, high reasoning effort | Sep 2025 | Official leaderboard |
| 11 | | 8.58 | agentic, high reasoning effort | Dec 2025 | Official leaderboard |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: Official QuantSightBench leaderboard
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official QuantSightBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
More in Forecasting
- ForecastBench In use Models give probabilities for 500 auto-generated questions per round about future events, drawn from prediction markets and time-series datasets. Forecasts are scored with difficulty-adjusted Brier scores once the questions resolve. GPT-5.5 0.11
- Prophet Arena In use Models and agents forecast open Kalshi prediction-market events and are scored once the events resolve. The live board ranks a difficulty-adjusted Arena score and also reports calibration error and market return. Claude Opus 5.5 86.2
- Guesswork (The Aggregate) Not in model cards Models predict benchmark scores before publication. The maker ranks forecasts by mean absolute error after scaling each error by the variation in that benchmark’s scores. GPT-6 Astra 0.56
- AGI-Pilled Bench Not in model cards Models forecast when AI will reach milestones that already happened in 2026. The score is the median years after 2026 across the forecast tasks. Claude Fable 5.2 6.50
- Dutch Book Bench Not in model cards Models assign probabilities to related events in separate conversations. The grader checks whether inconsistent probabilities let a bettor guarantee a profit.
Gemini 3.8 Flash 89.2%
- Echo General AI Prediction Leaderboard Not in model cards Models forecast unresolved events drawn from prediction markets, synthesized news questions and expert-written questions. Forecasts made at aligned times are compared in pairwise battles, and an Elo fit ranks the models.
DeepSeek-V4-Pro 1,018
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 8 months ago, in March 2026, for GPT-5.4. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the official leaderboard (11 entries).
- Pass Models scored 11 models we track.
- Warning Newest score The newest came 8 months ago, in March 2026, for GPT-5.4.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on quantsightbench.com: 11 rows for models we track at the last read.
- Pass Last read November 19th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.