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: 1,007 resolved forecasting questions — 759 binary and 248 numeric — researched and forecast against a frozen web corpus. Paper and dataset to follow.

BTF-3 Leaderboard

Evaluated: June 2026

All scores are on the Brier scale; lower is better, and the best score in each column is bolded.

Agent
Pooled score(n=1,007)
Binary(Brier, n=759)
Numeric(RPS, n=248)
1FutureSearch SOTA*
0.119 [0.1090.128]0.114 [0.1000.128]0.123 [0.1100.137]
2GPT-5.5 (agent SDK)
0.125 [0.1170.134]0.129 [0.1180.140]0.122 [0.1090.135]
3Claude Fable 5 (high)
0.127 [0.1170.138]0.124 [0.1090.140]0.130 [0.1170.144]
4Claude Opus 4.8 (xhigh)
0.128 [0.1180.138]0.126 [0.1120.141]0.130 [0.1170.143]
5Claude Opus 4.8 (agent SDK)
0.138 [0.1280.149]0.132 [0.1170.147]0.145 [0.1310.160]
6Claude Opus 4.8 (high)
0.140 [0.1300.150]0.128 [0.1140.143]0.152 [0.1380.165]
7GPT-5.5 (high)
0.140 [0.1310.149]0.140 [0.1280.152]0.141 [0.1270.154]
8Claude Opus 4.6
0.141 [0.1310.152]0.134 [0.1200.149]0.148 [0.1330.163]
9Claude Opus 4.7
0.145 [0.1340.157]0.134 [0.1200.150]0.157 [0.1400.174]
10Claude Sonnet 5 (xhigh)
0.150 [0.1400.160]0.147 [0.1330.162]0.153 [0.1390.167]
11Gemini 3.5 Flash
0.155 [0.1440.166]0.156 [0.1410.172]0.154 [0.1380.170]

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. † Run with an earlier version of our forecasting agent. ‡ Self-driving run via the model vendor's agent SDK (Claude Agent SDK / OpenAI Agents SDK) instead of our forecasting agent. Claude Fable 5 (high) is missing 23 binary questions (n=736) and 3 numeric questions (n=245). Claude Opus 4.7 is missing one numeric question (n=247).

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×)

1FutureSearch SOTA
2GPT-5.5 (agent SDK)
3Claude Fable 5 (high)
4Claude Opus 4.8 (xhigh)
5Claude Opus 4.8 (agent SDK)
6Claude Opus 4.8 (high)
7GPT-5.5 (high)
8Claude Opus 4.6
9Claude Opus 4.7
10Claude Sonnet 5 (xhigh)
11Gemini 3.5 Flash
1FutureSearch SOTA
-.007**
-.008**
-.009***
-.020***
-.021***
-.022***
-.022***
-.026***
-.031***
-.036***
2GPT-5.5 (agent SDK)
.007**
-.002
-.003
-.013**
-.014***
-.015***
-.016***
-.020***
-.025***
-.029***
3Claude Fable 5 (high)
.008**
.002
-.001
-.011***
-.012***
-.014***
-.014***
-.019***
-.024***
-.028***
4Claude Opus 4.8 (xhigh)
.009***
.003
.001
-.010***
-.012***
-.012***
-.013***
-.017***
-.022***
-.027***
5Claude Opus 4.8 (agent SDK)
.020***
.013**
.011***
.010***
-.002
-.002
-.003
-.007*
-.012**
-.016**
6Claude Opus 4.8 (high)
.021***
.014***
.012***
.012***
.002
.000
-.001
-.006
-.010**
-.015***
7GPT-5.5 (high)
.022***
.015***
.014***
.012***
.002
.000
-.001
-.005
-.010*
-.015***
8Claude Opus 4.6
.022***
.016***
.014***
.013***
.003
.001
.001
-.004
-.009*
-.014**
9Claude Opus 4.7
.026***
.020***
.019***
.017***
.007*
.006
.005
.004
-.005
-.010
10Claude Sonnet 5 (xhigh)
.031***
.025***
.024***
.022***
.012**
.010**
.010*
.009*
.005
-.005
11Gemini 3.5 Flash
.036***
.029***
.028***
.027***
.016**
.015***
.015***
.014**
.010
.005

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 Agent0.1190.0020.081
Opus 4.6 Agent0.1300.0050.075
Gemini 3.1 Pro Agent0.1410.0120.069
GPT-5.4 Agent0.1520.0100.056
Grok 4.20 Beta Agent0.1650.0030.039

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

Loading radar chart...
No data available

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:

TournamentOur standingLeader
Summer 2026 FutureEval Bot Tournamentlive#2 of 150acm_bot (605.55)
MiniBenchlivenot yet scored
MiniBench - 2026-06-15#1 of 118FutureSearch (1267.95)
MiniBench - 2026-06-01#2 of 113laertes (740.66)
MiniBench - 2026-05-18#3 of 114Preseen-Chestnut (1142.57)
MiniBench - 2026-05-04#7 of 119mmBot (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 #13 of 261 submitted models, at a Brier Index of 64.2.

Preliminary leaderboard

Updated: 2026-07-05

Brier Index; higher is better. Showing the top 15 of 261 models.

ModelBrier Index(95% CI)N
1
Torchcast AIrice-demon
67.5 [65.369.8]239
2
Torchcast AIcaptain-jack / dragon-brother
67.3 [65.169.6]239
4
Google DeepMindsilver-anchor
66.0 [63.668.7]239
5
Voicetreevoicetree-axiom-2
65.4 [63.767.1]239
6
Google DeepMindceramic-kettle
65.3 [62.867.9]239
7
Google DeepMindblue croc
65.1 [63.067.5]239
8
Google DeepMindbig green leaf / fire hedgehog
65.0 [62.967.5]239
10
Voicetreevoicetree-axiom-0
64.9 [63.366.7]239
11
Voicetreevoicetree-axiom-1
64.6 [62.966.4]239
12
Google DeepMindgreen tree
64.4 [63.365.4]727
13
FutureSearchfb_early_closer_v2 / fb_forecaster_v2
64.2 [62.566.3]239
15
Superforecaster median forecastForecastBench
63.6 [62.364.8]521
16
Google DeepMindiron-compass
63.3 [60.366.1]239
17
Google DeepMindyellow mouse
63.0 [62.164.0]727
18
xAIGrok 4.20 (Beta, D)
62.7 [61.564.0]439

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:

Diagram showing how RetroSearch provides frozen web snapshots to agents
Illustration of the system architecture of Deep Research Bench using RetroSearch. This shows the flow from task definition through the scraping pipeline that populates the RetroSearch database prior to running the benchmark, and then how agents use RetroSearch via an API at the time of task evaluation.