AI benchmark tape
Benchmark Live AI capital allocation same brief, frozen portfolios, real returns AI Positioning Risk-seeking 72.5/100 As of September 21, 2026. Broad risk seeking Model Agreement Tight 3.5 point dispersion Insight Models missed the monthly oracle asset The hindsight best asset was Crude Oil (USO) at +16.47%. 0 of 7 models held it, with +0.00% average allocation. Crowded Position S&P 500 23.9% 6 models hold SPY Biggest Shift Japan Equities up 20.0 pp vs prior weekly round Open Portfolio Leader Claude Fable 5.1 currently leads at 1.6% interim result, not yet scored Latest Weekly Grok 4.6 led 0.6% vs S&P 0.6% Full Weekly History Grok 4.5 leads score 6.2 across 44/71
Data Prices Sep 25 146/146 audited

Weekly comparison set

Weekly Set: Aug 13, 2026

Weekly comparison set automatically opened when the Aug 13 official roster first required a new equal-run benchmark group across 8 models.

Benchmark Forming

Fairness Scope

Every ranked model in this set is scored only on rounds that all 8 listed models completed. If one model misses a resolved round, that round is excluded from this set for everyone.

All comparison sets
Shared rounds3 Models8 Threshold6 StatusBenchmark Forming
Equal-run benchmark

Weekly Benchmark Forming

Every ranked model in this set completed the same 3 weekly rounds.

3 shared resolved rounds8 equal-run models ranked3 more shared rounds to qualifyNewest included round: CB-2026-08-18-1W
Shared resolved rounds

CapitalBench Score

A score of 30 means the model earned 30% of the best possible return across these rounds. Calculation

Grok 4.3
Claude Fable 5
GPT-5.6 Sol
Claude Opus 5
Gemini 3.1 Pro
Grok 4.5
Claude Opus 4.8
Grok 4.6
S&P 500
Max possible What is this? Max possible is the best eligible asset after scoring for the same rounds. It is a hindsight ceiling, not a model portfolio. hindsight best asset

A score of 30 means the model earned 30% of the best possible return across these rounds. Calculation

Grok 4.3 xAI · 3/3 scored rounds
8.6
Claude Fable 5 Anthropic · 3/3 scored rounds
5.7
GPT-5.6 Sol OpenAI · 3/3 scored rounds
4.8
Claude Opus 5 Anthropic · 3/3 scored rounds
4.8
Gemini 3.1 Pro Google · 3/3 scored rounds
4.3
Grok 4.5 xAI · 3/3 scored rounds
4.3
Claude Opus 4.8 Anthropic · 3/3 scored rounds
-0.0
Grok 4.6 xAI · 3/3 scored rounds
-2.0
S&P 500 S&P 500 · 3/3 scored rounds
-4.1
Max possible Hindsight ceiling, not a model portfolio
What is this? Max possible is the best eligible asset after scoring for the same rounds. It is a hindsight ceiling, not a model portfolio. 100.0
3 shared resolved rounds8 equal-run models ranked3 more shared rounds to qualifyNewest included round: CB-2026-08-18-1W
Return context

Average Return Details

Average portfolio return across the same finished rounds.

xAI Grok 4.3
2.35%
Anthropic Claude Fable 5
1.55%
OpenAI GPT-5.6 Sol
1.31%
Anthropic Claude Opus 5
1.30%
Google Gemini 3.1 Pro
1.18%
xAI Grok 4.5
1.17%
Anthropic Claude Opus 4.8
-0.01%
xAI Grok 4.6
-0.53%
S&P S&P 500
-1.12%
MAX Max possible What is this? Max possible is the best eligible asset after scoring for the same rounds. It is a hindsight ceiling, not a model portfolio.
27.24%
Excluded for fairness: CB-2026-08-19-1W missing anthropic-claude-opus-4-8; CB-2026-08-20-1W missing anthropic-claude-opus-4-8; CB-2026-08-21-1W missing anthropic-claude-opus-4-8; CB-2026-08-23-1W missing anthropic-claude-opus-4-8; CB-2026-08-24-1W missing anthropic-claude-opus-4-8; CB-2026-08-25-1W missing anthropic-claude-opus-4-8; CB-2026-08-26-1W missing anthropic-claude-opus-4-8; CB-2026-08-27-1W missing anthropic-claude-opus-4-8; CB-2026-08-30-1W missing anthropic-claude-opus-4-8; CB-2026-09-01-1W missing anthropic-claude-opus-4-8; CB-2026-09-02-1W missing anthropic-claude-opus-4-8; CB-2026-09-03-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8; CB-2026-09-04-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-09-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-10-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-11-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-13-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-15-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-16-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol Fairness rule: every ranked model completed every included round. A missed round is excluded from this set for everyone. Forming: this set becomes the Current Weekly Benchmark at 6 shared resolved rounds.
Risk and return

Who earned more return for the risk they took?

Average realized return and frozen portfolio risk across the same 3 shared weekly rounds.

Risk and return for models in this comparison setAverage frozen portfolio allocation risk is plotted horizontally and average realized return is plotted vertically across 3 shared rounds.-2.00%0.00%2.00%4.00%5060708090CapitalBench allocation riskAverage returnS&P 500Beat S&P with lower allocation riskGrok 4.3Fable 5GPT-5.6 SolOpus 5Gemini 3.1Grok 4.5Opus 4.8Grok 4.6
Risk and return for models in this comparison setAverage frozen portfolio allocation risk is plotted horizontally and average realized return is plotted vertically across 3 shared rounds.-2.00%0.00%2.00%4.00%5060708090CapitalBench allocation riskAverage returnS&P 500Grok 4.3Fable 5GPT-5.6 SolOpus 5Gemini 3.1Grok 4.5Opus 4.8Grok 4.6
Grok 4.3xAI
+2.35%average return72.2/100Risk-seeking+3.47 ppversus S&P 50052.9-87.5risk range

Return leaderGrok 4.3 led the models at +2.35% average return with a 72.2/100 risk score.

Benchmark testClaude Opus 4.8 beat the S&P 500 while taking no more allocation risk.
Compare model groups

How do these results compare?

Grok 4.5 ranks first in Sep 4 Weekly. Grok 4.3 ranks first in Aug 13 Weekly. The groups have no completed rounds in common. Sep 4 Weekly includes 7 more rounds, while Aug 13 Weekly includes 3 more rounds. Claude Fable 5, GPT-5.6 Sol, Claude Opus 4.8 appear only in Aug 13 Weekly. Claude Fable 5.1, GPT-6 Astra appear only in Sep 4 Weekly.

5models in both 0rounds used by both Changed a lotchange in order Yes top model changed

Use Sep 4 Weekly as the more reliable ranking because it has 7 completed rounds. Aug 13 Weekly has 3 and needs 3 more before it has enough evidence to become the main ranking.

Compare these groups
Roster

Models In This Set

This roster stays fixed so the set can keep growing as a clean equal-run comparison.

Anthropic Claude Fable 5

anthropic-claude-fable-5

3 shared rounds in this set Anthropic Claude Opus 4.8

anthropic-claude-opus-4-8

3 shared rounds in this set Anthropic Claude Opus 5

anthropic-claude-opus-5

3 shared rounds in this set Google Gemini 3.1 Pro

google-gemini-3-1-pro

3 shared rounds in this set OpenAI GPT-5.6 Sol

openai-gpt-5-6-sol

3 shared rounds in this set xAI Grok 4.3

xai-grok-4-3

3 shared rounds in this set xAI Grok 4.5

xai-grok-4-5

3 shared rounds in this set xAI Grok 4.6

xai-grok-4-6

3 shared rounds in this set
Round audit

Included And Excluded Rounds

Included rounds count toward the score. Excluded rounds are resolved rounds inside this comparison history where at least one set model was missing.

3 more to qualify
Included rounds CB-2026-08-13-1W, CB-2026-08-15-1W, CB-2026-08-18-1W
Excluded for fairness
19 resolved candidate rounds CB-2026-08-19-1W missing anthropic-claude-opus-4-8; CB-2026-08-20-1W missing anthropic-claude-opus-4-8; CB-2026-08-21-1W missing anthropic-claude-opus-4-8; CB-2026-08-23-1W missing anthropic-claude-opus-4-8; CB-2026-08-24-1W missing anthropic-claude-opus-4-8; CB-2026-08-25-1W missing anthropic-claude-opus-4-8; CB-2026-08-26-1W missing anthropic-claude-opus-4-8; CB-2026-08-27-1W missing anthropic-claude-opus-4-8; CB-2026-08-30-1W missing anthropic-claude-opus-4-8; CB-2026-09-01-1W missing anthropic-claude-opus-4-8; CB-2026-09-02-1W missing anthropic-claude-opus-4-8; CB-2026-09-03-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8; CB-2026-09-04-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-09-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-10-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-11-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-13-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-15-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol; CB-2026-09-16-1W missing anthropic-claude-fable-5, anthropic-claude-opus-4-8, openai-gpt-5-6-sol
Calculation

How The Score Is Calculated

CapitalBench Score equals total model return across included shared rounds divided by total max-possible return across those same rounds, multiplied by 100. Max possible is the best eligible asset in each included round in hindsight.

Scoring details