futuresearch evals
Want to run on our benchmarks? Please contact us at evals@futuresearch.ai.
How our forecasting and research agents perform on our own benchmarks, on live public leaderboards, and on real markets.
Bench to the Future 3 (BTF-3)
BTF-3 is the third edition of our pastcasting benchmark: 1,907 resolved forecasting questions — 1,515 binary and 392 numeric — researched and forecast against a frozen web corpus. Paper and dataset to follow.
BTF-3 Leaderboard
Evaluated: June–July 2026
All scores are on the Brier scale; lower is better, and the best score in each column is bolded.
Agent | Pooled score(n=1,907) | Binary(Brier, n=1,515) | Numeric(RPS, n=392) |
|---|---|---|---|
1FutureSearch SOTA* | 0.122 [0.115–0.129] | 0.120 [0.110–0.130] | 0.124 [0.114–0.135] |
2Claude Opus 4.8 (xhigh) | 0.130 [0.123–0.138] | 0.131 [0.121–0.142] | 0.129 [0.119–0.139] |
3Claude Fable 5 (high) | 0.131 [0.123–0.138] | 0.132 [0.122–0.143] | 0.129 [0.119–0.139] |
4GPT-5.5 (high, agent SDK)‡ | 0.134 [0.128–0.140] | 0.142 [0.134–0.150] | 0.124 [0.114–0.135] |
5GPT-5.6 Sol (high) | 0.135 [0.128–0.143] | 0.141 [0.132–0.150] | 0.129 [0.117–0.141] |
6Claude Opus 4.8 (high, agent SDK)‡ | 0.137 [0.129–0.145] | 0.135 [0.124–0.146] | 0.140 [0.130–0.151] |
7Claude Opus 4.8 (high) | 0.140 [0.132–0.147] | 0.135 [0.125–0.146] | 0.145 [0.134–0.157] |
8GPT-5.5 (high) | 0.143 [0.136–0.149] | 0.148 [0.140–0.156] | 0.136 [0.125–0.147] |
9Claude Sonnet 5 (xhigh) | 0.154 [0.146–0.162] | 0.154 [0.144–0.164] | 0.154 [0.143–0.166] |
Binary questions are scored by the Brier score (mean squared error of the forecast probability), numeric questions by a normalized ranked probability score (RPS), which generalizes the Brier score to distributional forecasts. The pooled score averages across all questions, counting each numeric question three times as much as a binary one (numeric forecasts are more informative per question).
Brackets are 95% confidence intervals, computed by percentile bootstrap (5,000 resamples of the question set).
* FutureSearch SOTA synthesizes forecasts from multiple FutureSearch agent runs. ‡ Self-driving run via the model vendor's agent SDK (Claude Agent SDK / OpenAI Agents SDK) instead of our forecasting agent. FutureSearch SOTA is missing 88 binary questions (n=1,427) and 9 numeric questions (n=383). Claude Fable 5 (high) is missing 29 binary questions (n=1,486) and 3 numeric questions (n=389). GPT-5.5 (high, agent SDK) is missing 50 binary questions (n=1,465) and one numeric question (n=391). Claude Opus 4.8 (high, agent SDK) is missing 8 binary questions (n=1,507).
CHAMPS KNOW strategic emphasis
Mean Borda score per dimension (rank 1 = 10 … rank 10 = 1); higher means the dimension is more prominent in the agent's rationales. The top 3 agents are shown by default — click a name to add or remove it.
Pairwise comparisons
Paired bootstrap on pooled scores (numeric weighted 3×)
Each cell is the difference in pooled score (row − column) on the questions both agents forecast; negative (green) means the row agent is more accurate. Bold, bordered cells are statistically significant (two-sided paired-bootstrap * p<.05, ** p<.01, *** p<.001); grey cells are not. Hover a cell for the 95% confidence interval, p-value, and shared question count.
Bench to the Future 2 (BTF-2)
BTF-2 evaluates agents on 1,417 hard forecasting questions. Agents research and forecast offline against a frozen 15M-document corpus. Rationales and reasoning traces are evaluated for strategic reasoning.
BTF-2 Leaderboard
Last updated: 2026-04-20
Agent | Brier (accuracy) | Calibration | Refinement |
|---|---|---|---|
| FutureSearch Agent | 0.119 | 0.002 | 0.081 |
| Opus 4.6 Agent | 0.130 | 0.005 | 0.075 |
| Gemini 3.1 Pro Agent | 0.141 | 0.012 | 0.069 |
| GPT-5.4 Agent | 0.152 | 0.010 | 0.056 |
| Grok 4.20 Beta Agent | 0.165 | 0.003 | 0.039 |
Brier scores on 1,417 pastcasting questions (lower is better). The FutureSearch Agent is an ensemble significantly more accurate than any single frontier agent. Radar chart shows CHAMPS KNOW strategic emphasis (Borda scores, 8 of 10 dimensions).
Papers
Datasets
Deep Research Bench (DRB)
DRB benchmarks how well LLM agents do research on the web. Each of the 0 diverse, real-world tasks provides 10-100k webpages stored offline for search and reasoning, accompanied by carefully curated answers.
DRB Leaderboard
Last updated:
Agent | Score | Cost ($) | Runtime (s) |
|---|---|---|---|
Scores averaged first per task category (radar chart), then across all tasks (table). Runtime is estimated from ReAct steps, not wall-clock time.
Papers
Metaculus AI Forecasting Tournaments
Metaculus runs live bot tournaments where forecasting agents predict real, unresolved questions and are scored against the field by spot peer score. Our standing in the tournaments we take part in, refreshed at each deploy:
| Tournament | Our standing | Leader |
|---|---|---|
| Summer 2026 FutureEval Bot Tournamentlive | #1 of 163 | FutureSearch (927.39) |
| MiniBenchlive | not yet scored | — |
| MiniBench - 2026-06-15 | #1 of 118 | FutureSearch (1267.95) |
| MiniBench - 2026-06-01 | #2 of 113 | laertes (740.66) |
| MiniBench - 2026-05-18 | #3 of 114 | Preseen-Chestnut (1142.57) |
| MiniBench - 2026-05-04 | #7 of 119 | mmBot (1766.40) |
Standings are pulled from the Metaculus API at deploy time. Bots are scored by spot peer score (a per-question comparison against every other forecaster on the same question); higher is better. The leader column shows the top-ranked bot and its score for context. MiniBench tournaments run on a rolling two-week cadence.
ForecastBench
ForecastBench is a dynamic, contamination-free benchmark of AI forecasting accuracy run by the Forecasting Research Institute. Bots forecast hundreds of unresolved real-world questions, scored on a Brier Index (0–100, higher is better). FutureSearch's forecasting agent currently ranks #14 of 273 submitted models, at a Brier Index of 63.9.
Preliminary leaderboard
Updated: 2026-07-11
Brier Index; higher is better. Showing the top 15 of 273 models.
| Model | Brier Index(95% CI) | N |
|---|---|---|
1 Torchcast AIrice-demon | 65.9 [64.5–67.5] | 426 |
2 Torchcast AIcaptain-jack | 65.6 [64.1–67.2] | 426 |
3 Torchcast AIdragon-brother | 65.5 [64.0–67.0] | 426 |
3 Voicetreevoicetree-axiom-2 | 65.5 [64.3–66.7] | 426 |
5 Google DeepMindsilver-anchor | 65.3 [63.4–67.2] | 426 |
6 Google DeepMindbig green leaf / blue croc / fire hedgehog | 65.1 [63.5–66.8] | 426 |
6 Voicetreevoicetree-axiom-0 | 65.1 [63.9–66.3] | 426 |
10 Google DeepMindceramic-kettle | 64.6 [62.7–66.5] | 426 |
10 Voicetreevoicetree-axiom-1 | 64.6 [63.4–65.8] | 426 |
12 Google DeepMindgreen tree | 64.2 [63.1–65.2] | 727 |
13 Torchcast AIcarb-bomb-nano | 64.0 [62.0–66.0] | 244 |
14 FutureSearchfb_early_closer_v2 / fb_forecaster_v2 | 63.9 [62.6–65.2] | 426 |
14 Google DeepMindiron-compass | 63.9 [61.8–65.8] | 426 |
17 Torchcast AIwyrm-warlord-nano | 63.8 [61.8–65.7] | 244 |
18 Superforecaster median forecastForecastBench | 63.7 [62.4–65.0] | 521 |
The preliminary leaderboard ranks models on questions from the current dataset that have already resolved; it is regenerated nightly from the public dataset repository. The Brier Index rescales the mean Brier score to a 0–100 scale (100 = perfect, 50 = uninformed) and adjusts for question difficulty. Brackets are 95% confidence intervals; N is the number of resolved questions scored.
RetroSearch
DRB and BTF-2 use RetroSearch, a system designed to serve agents a frozen, previously scraped version of the internet instead of the live pages, allowing reproducible runs even as the internet changes, and enabling forecasting tasks to be run as "pastcasting".
RetroSearch aims to emulate Google search (specifically, the Serper search API) as closely as possible, so as to minimize differences between live and "retro" agent runs. A single RetroSearch search query follows the following steps:
- Run a live Serper search for the query
- Look up pages obtained from live search in the RetroSearch database and other archive sources
- If the page is not found in the RetroSearch database, remove it from the results
- Write new snippets from a sample of page content using a simple LLM
- Return the results in the original format of the Google results
This approach ensures a search experience for agents that is consistent with real search, but backed exclusively by pages we have a frozen candidate for. The following diagram from the paper illustrates the process:
