AI benchmark tape
Benchmark Live AI capital allocation same brief, frozen portfolios, real returns AI Positioning Aggressive 89.5/100 As of July 10, 2026. Growth-led risk seeking Model Agreement Tight 3.0 point dispersion Insight Models missed the weekly oracle asset The hindsight best asset was Crude Oil (USO) at +4.84%. 0 of 6 models held it, with +0.00% average allocation. Crowded Position Semiconductors 18.8% 8 models hold SMH Biggest Shift Cybersecurity down 11.6 pp vs prior weekly round Live MTM Grok 4.5 -0.3% not final Latest Weekly Gemini 3.1 Pro led 0.2% vs S&P 0.9% Full Weekly History Claude Fable 5 leads score 16.3 across 5/25
Data Prices Jul 9 61/61 audited
CapitalBench benchmark

Market Regime Benchmark

A market regime is simply a type of market move: falling, flat, or rising. See which AI model portfolios do best in each one. Every market type gets the same importance in the main ranking.

Updated July 9, 2026 Monthly and weekly views Completed tests only
How this page works
Models are ranked byAverage return across market types
Models are compared withS&P 500
A model gets a rank whenIt has enough results in every market type used
Monthly results selected
Quick answer

Who leads, and did they beat the S&P 500?

Start here for the result. The charts and exact numbers below explain it.

Which model leads?

Claude Opus 4.8

Anthropic

+0.41%

Average return across every market type with enough data.

Did the leader beat the S&P 500?

Yes

+0.34 points

S&P 500 average return across the same market types: +0.07%

How much data supports this?

Some evidence

2 of 5

2 market types have enough matching results. Each ranked model contributes 7 results.

Model ranking

Which model performs best across different markets?

Models are ranked by their average return across the market types that have enough data.

Models are ranked byAverage return across market typesHigher is better
Market types used2 of 5Enough results to compare
Models ranked50 need more results
Results used7 per modelSame dates for every ranked model
Main ranking

Who had the highest average return?

Bars to the right are gains. Bars to the left are losses.

Claude Opus 4.8 Anthropic · 2/2 market types +0.34 points vs S&P
Claude Opus 4.7 Anthropic · 2/2 market types +0.33 points vs S&P
Grok 4.3 xAI · 2/2 market types +0.13 points vs S&P
Gemini 3.1 Pro Google · 2/2 market types -1.93 points vs S&P
GPT-5.5 OpenAI · 2/2 market types -3.51 points vs S&P
S&P S&P 500 Comparison S&P 500 return
Grok 4.3 xAI · 2/2 market types
+0.20% +0.13 points vs S&P
Gemini 3.1 Pro Google · 2/2 market types
-1.86% -1.93 points vs S&P
GPT-5.5 OpenAI · 2/2 market types
-3.44% -3.51 points vs S&P
S&P S&P 500 Comparison
+0.07% S&P 500 return
  • Claude Opus 4.8: +0.41%. +0.34 points vs S&P
  • Claude Opus 4.7: +0.40%. +0.33 points vs S&P
  • Grok 4.3: +0.20%. +0.13 points vs S&P
  • Gemini 3.1 Pro: -1.86%. -1.93 points vs S&P
  • GPT-5.5: -3.44%. -3.51 points vs S&P
  • S&P 500: +0.07%. S&P 500 return
Score view

How did each model score?

100 is the best possible result after the fact. It is a reference, not a real portfolio.

Grok 4.3 xAI
-0.5 Rank 3 · +0.20% return
GPT-5.5 OpenAI
-17.1 Rank 5 · -3.44% return
S&P S&P 500 Comparison
-1.0 Same market-type weights
BEST Best possible Known only after prices changed
100.0 Reference only
  • Claude Opus 4.8: 0.4. Rank 1 · +0.41% return
  • Claude Opus 4.7: 0.1. Rank 2 · +0.40% return
  • Grok 4.3: -0.5. Rank 3 · +0.20% return
  • Gemini 3.1 Pro: -10.6. Rank 4 · -1.86% return
  • GPT-5.5: -17.1. Rank 5 · -3.44% return
  • S&P 500: -1.0. Same market-type weights
  • Best possible: 100.0. Reference only
S&P

S&P 500 comparison: +0.07% average return and -1.0 CapitalBench Score across the same market types.

2 market types used
Exact numbers

What are the full monthly results?

Rank follows average return. The other columns show how the result was built.

Rank and model Average return Difference from S&P CapitalBench Score Market types Results used Beat S&P
1 Claude Opus 4.8Anthropic +0.41% +0.34 points 0.4 2/2 7 57.14%
2 Claude Opus 4.7Anthropic +0.40% +0.33 points 0.1 2/2 7 71.43%
3 Grok 4.3xAI +0.20% +0.13 points -0.5 2/2 7 71.43%
4 Gemini 3.1 ProGoogle -1.86% -1.93 points -10.6 2/2 7 0.00%
5 GPT-5.5OpenAI -3.44% -3.51 points -17.1 2/2 7 28.57%
Falling and rising markets

Who does best when markets fall or rise?

Each chart uses the same dates for every model shown, with the S&P 500 beside them.

Market fell

When the S&P 500 fell

S&P < -1.0%

Some evidence

Grok 4.3 ranks first here with a -1.27% return, beating the S&P 500 by 0.70 points.

Grok 4.3 xAI · 5 results
-1.27% +0.70 points vs S&P
GPT-5.5 OpenAI · 5 results
-4.17% -2.20 points vs S&P
S&P S&P 500 5 matching dates
-1.97% S&P 500 return
  • Grok 4.3: -1.27%. +0.70 points vs S&P
  • Claude Opus 4.8: -1.33%. +0.65 points vs S&P
  • Claude Opus 4.7: -1.80%. +0.18 points vs S&P
  • Gemini 3.1 Pro: -3.87%. -1.89 points vs S&P
  • GPT-5.5: -4.17%. -2.20 points vs S&P
  • S&P 500: -1.97%. S&P 500 return

Both charts use the same scale, with zero in the middle.

Show exact numbers and ranks
ModelReturnvs S&P
S&P S&P 500 -1.97% Comparison
1 Grok 4.3 -1.27% +0.70 points
2 Claude Opus 4.8 -1.33% +0.65 points
3 Claude Opus 4.7 -1.80% +0.18 points
4 Gemini 3.1 Pro -3.87% -1.89 points
5 GPT-5.5 -4.17% -2.20 points
Dates used
5 of 7
Models compared
5
Leader repeated
60%
Lead over #2
+0.05 points
Market rose

When the S&P 500 rose

S&P > +1.0%

Some evidence

Claude Opus 4.7 ranks first here with a +2.79% return, beating the S&P 500 by 0.97 points.

Grok 4.3 xAI · 3 results
+2.20% +0.38 points vs S&P
GPT-5.5 OpenAI · 3 results
-1.71% -3.53 points vs S&P
S&P S&P 500 3 matching dates
+1.82% S&P 500 return
  • Claude Opus 4.7: +2.79%. +0.97 points vs S&P
  • Claude Opus 4.8: +2.50%. +0.68 points vs S&P
  • Grok 4.3: +2.20%. +0.38 points vs S&P
  • Gemini 3.1 Pro: +0.93%. -0.89 points vs S&P
  • GPT-5.5: -1.71%. -3.53 points vs S&P
  • S&P 500: +1.82%. S&P 500 return

Both charts use the same scale, with zero in the middle.

Show exact numbers and ranks
ModelReturnvs S&P
S&P S&P 500 +1.82% Comparison
1 Claude Opus 4.7 +2.79% +0.97 points
2 Claude Opus 4.8 +2.50% +0.68 points
3 Grok 4.3 +2.20% +0.38 points
4 Gemini 3.1 Pro +0.93% -0.89 points
5 GPT-5.5 -1.71% -3.53 points
Dates used
3 of 3
Models compared
5
Leader repeated
67%
Lead over #2
+0.29 points
Market types

How did each model perform in each type of market?

The table shows model return first, followed by the difference from the S&P 500.

Large fall Weekly: <= -2.0% Monthly: <= -3.0% 4 weekly / 1 monthly results
Small fall Weekly: -2.0% to -0.5% Monthly: -3.0% to -1.0% 4 weekly / 6 monthly results
Little change Weekly: -0.5% to +0.5% Monthly: -1.0% to +1.0% 4 weekly / 1 monthly results
Small rise Weekly: +0.5% to +2.0% Monthly: +1.0% to +3.0% 11 weekly / 3 monthly results
Large rise Weekly: >= +2.0% Monthly: >= +3.0% 2 weekly / 0 monthly results
Model Large fall S&P <= -3.0% 0/1 dates used Small fall -3.0% to -1.0% 4/6 dates used Little change -1.0% to +1.0% 0/1 dates used Small rise +1.0% to +3.0% 3/3 dates used Large rise S&P >= +3.0% 0/0 dates used
S&PS&P 500Comparison -3.15% all available dates -1.68% same dates +0.24% all available dates +1.82% same dates n/a all available dates
1 Claude Opus 4.82/2 market types n/aNot enough matching results -1.68% 0.00 points vs S&P n/aNot enough matching results +2.50% +0.68 points vs S&P n/aNot enough matching results
2 Claude Opus 4.72/2 market types n/aNot enough matching results -1.99% -0.31 points vs S&P n/aNot enough matching results +2.79% +0.97 points vs S&P n/aNot enough matching results
3 Grok 4.32/2 market types n/aNot enough matching results -1.80% -0.12 points vs S&P n/aNot enough matching results +2.20% +0.38 points vs S&P n/aNot enough matching results
4 Gemini 3.1 Pro2/2 market types n/aNot enough matching results -4.66% -2.98 points vs S&P n/aNot enough matching results +0.93% -0.89 points vs S&P n/aNot enough matching results
5 GPT-5.52/2 market types n/aNot enough matching results -5.17% -3.49 points vs S&P n/aNot enough matching results -1.71% -3.53 points vs S&P n/aNot enough matching results
Large fallS&P <= -3.0% 0/1used / available Early result

1 result is available. 2 more needed before models can be compared here.

Small fall-3.0% to -1.0% 4/6used / available Some evidence

Claude Opus 4.8 ranks first among the models and matches the S&P 500.

Grok 4.3 xAI · 4 results
-1.80% -0.12 points vs S&P
GPT-5.5 OpenAI · 4 results
-5.17% -3.49 points vs S&P
S&P S&P 500 4 matching dates
-1.68% S&P 500 return
  • Claude Opus 4.8: -1.68%. 0.00 points vs S&P
  • Grok 4.3: -1.80%. -0.12 points vs S&P
  • Claude Opus 4.7: -1.99%. -0.31 points vs S&P
  • Gemini 3.1 Pro: -4.66%. -2.98 points vs S&P
  • GPT-5.5: -5.17%. -3.49 points vs S&P
  • S&P 500: -1.68%. S&P 500 return

Every monthly market-type chart uses this same scale.

Models compared
5
Leader repeated
25%
Lead over #2
+0.11 points
Estimated range for #1
-4.64% to +1.28%
Little change-1.0% to +1.0% 0/1used / available Early result

1 result is available. 2 more needed before models can be compared here.

Small rise+1.0% to +3.0% 3/3used / available Some evidence

Claude Opus 4.7 ranks first here with a +2.79% return, beating the S&P 500 by 0.97 points.

Grok 4.3 xAI · 3 results
+2.20% +0.38 points vs S&P
GPT-5.5 OpenAI · 3 results
-1.71% -3.53 points vs S&P
S&P S&P 500 3 matching dates
+1.82% S&P 500 return
  • Claude Opus 4.7: +2.79%. +0.97 points vs S&P
  • Claude Opus 4.8: +2.50%. +0.68 points vs S&P
  • Grok 4.3: +2.20%. +0.38 points vs S&P
  • Gemini 3.1 Pro: +0.93%. -0.89 points vs S&P
  • GPT-5.5: -1.71%. -3.53 points vs S&P
  • S&P 500: +1.82%. S&P 500 return

Every monthly market-type chart uses this same scale.

Models compared
5
Leader repeated
67%
Lead over #2
+0.29 points
Estimated range for #1
+1.96% to +3.62%
Large riseS&P >= +3.0% 0/0used / available Early result

No completed tests fall into this market type yet.

What the results show

What patterns stand out?

Each statement shows the result and how much data supports it.

Monthly · when the market fellEnough data

Grok 4.3 beat the S&P 500 when the market fell.

Grok 4.3 ranked first among the models. Its difference from the S&P 500 was +0.70 points.

-1.27%return
S&P 500
-1.97%
Used / available
5/7
Leader repeated
60%
Monthly · when the market was flatEarly result

We need more results when the market was flat.

1 result is available, but not enough dates were completed by every model.

n/areturn
Dates used
0
Dates available
1
S&P 500 move
-1.0% to +1.0%
Monthly · when the market roseEnough data

Claude Opus 4.7 beat the S&P 500 when the market rose.

Claude Opus 4.7 ranked first among the models. Its difference from the S&P 500 was +0.97 points.

+2.79%return
S&P 500
+1.82%
Used / available
3/3
Leader repeated
67%
Show more patterns across market types 2
Monthly · across market typesEnough data

Grok 4.3 has the best worst-case score across the tested market moves.

Its lowest CapitalBench Score is -7.1 across 2 kinds of market move.

+0.47%average return
S&P 500
-0.08%
Average score
0.7
Market moves
2/2
Monthly · biggest changeEnough data

Gemini 3.1 Pro changes the most between falling and rising markets.

Its average return moves from -3.87% when the market fell to +0.93% when it rose.

4.80 pointsreturn difference
Falling-market score
-21.7
Rising-market score
3.6
Results used
5/3
Check the work

How can you check these results?

The method, a second ranking check, and every included test are available below.

Simple methodHow are the results calculated?Same dates, clear market groups, and equal importance
1

Use completed tests

Open tests do not count. Only finished results are included.

2

Group each market move

Each test is placed in a market type using the S&P 500 return over the same dates.

3

Compare the same dates

A chart uses only dates completed by every model shown in that chart.

4

Give each type equal importance

A market type needs at least 3 results before it affects the main ranking.

Average return
The model portfolio's average gain or loss on the dates shown.
Difference from S&P
Model return minus S&P 500 return on the same dates.
Average across market types
First average each market type, then average those results so every type matters equally.
CapitalBench Score
A score that compares the model with the S&P 500 and the best available choice after the fact. See the formula.
Beat-S&P rate
The percentage of included tests where the model did better than the S&P 500.
Leader repeated
How often the same model stays first when one date is removed and the ranking is recalculated.

Weekly and monthly periods can overlap, so do not add them together as if they were separate trades. Result-strength labels describe how much data supports a comparison; they do not promise what will happen next.

Second checkDoes the ranking change with equal sample sizes?Uses the same number of falling, flat, and rising market results
Monthly second check

More results are needed for this check

Not enough data
Down 7 Enough results
Flat 1 2 more needed
Up 3 Enough results

This check needs at least 3 falling, 3 flat, and 3 rising market results before it can rank models.

Full listWhich completed tests are included?36 tests with filters and source links
Open the data

Showing 10 of 36 tests

TestViewDatesS&P returnMarket typeAvg. model differenceModels
CB-2026-07-02-1W Weekly 2026-07-02 to 2026-07-09 +0.93% Small rise -1.45 points 6
CB-2026-06-09-1M Monthly 2026-06-09 to 2026-07-09 +2.25% Small rise -0.17 points 6
CB-2026-07-01-1W Weekly 2026-07-01 to 2026-07-08 -0.05% Little change +0.89 points 6
CB-2026-06-08-1M Monthly 2026-06-08 to 2026-07-08 +1.09% Small rise -0.37 points 5
CB-2026-06-30-1W Weekly 2026-06-30 to 2026-07-07 +0.13% Little change -4.15 points 5
CB-2026-06-29-1W Weekly 2026-06-29 to 2026-07-06 +1.39% Small rise -2.30 points 5
CB-2026-06-05-1M Monthly 2026-06-05 to 2026-07-06 +2.12% Small rise -0.99 points 5
CB-2026-06-26-1W Weekly 2026-06-26 to 2026-07-02 +2.17% Large rise -0.78 points 5
CB-2026-06-25-1W Weekly 2026-06-25 to 2026-07-02 +1.43% Small rise -0.19 points 5
CB-2026-06-03-1M Monthly 2026-06-02 to 2026-07-02 -1.70% Small fall -5.10 points 5
CB-2026-06-24-1W Weekly 2026-06-24 to 2026-07-01 +1.71% Small rise -0.01 points 5
CB-2026-06-02-1M Monthly 2026-06-01 to 2026-07-01 -1.43% Small fall -1.17 points 5
CB-2026-06-23-1W Weekly 2026-06-23 to 2026-06-30 +1.80% Small rise +2.20 points 5
CB-2026-06-22-1W Weekly 2026-06-22 to 2026-06-29 -0.46% Little change -2.74 points 5
CB-2026-06-01-1M Monthly 2026-05-29 to 2026-06-29 -1.79% Small fall +0.85 points 5
CB-2026-05-29-1M Monthly 2026-05-29 to 2026-06-29 -1.79% Small fall -0.10 points 5
CB-2026-05-28-1M Monthly 2026-05-28 to 2026-06-26 -3.15% Large fall +2.95 points 5
CB-2026-06-18-1W Weekly 2026-06-18 to 2026-06-25 -1.67% Small fall -1.06 points 5
CB-2026-06-17-1W Weekly 2026-06-17 to 2026-06-24 -0.79% Small fall -0.19 points 5
CB-2026-05-24-1M Monthly 2026-05-22 to 2026-06-24 -1.41% Small fall +6.61 points 4
CB-2026-06-16-1W Weekly 2026-06-16 to 2026-06-23 -1.98% Small fall +0.33 points 5
CB-2026-06-15-1W Weekly 2026-06-15 to 2026-06-22 -1.13% Small fall +3.82 points 5
CB-2026-06-13-1W Weekly 2026-06-12 to 2026-06-18 +0.67% Small rise +3.28 points 6
CB-2026-06-12-1W Weekly 2026-06-11 to 2026-06-18 +1.22% Small rise -0.86 points 6
CB-2026-05-17-1M Monthly 2026-05-15 to 2026-06-17 +0.24% Little change -4.68 points 4
CB-2026-06-09-1W Weekly 2026-06-09 to 2026-06-16 +1.80% Small rise -1.52 points 6
CB-2026-06-08-1W Weekly 2026-06-08 to 2026-06-15 +2.11% Large rise -2.23 points 5
CB-2026-06-05-1W Weekly 2026-06-05 to 2026-06-12 +0.57% Small rise -0.43 points 5
CB-2026-05-10-1M Monthly 2026-05-08 to 2026-06-10 -1.65% Small fall +2.42 points 4
CB-2026-06-03-1W Weekly 2026-06-02 to 2026-06-09 -2.96% Large fall -5.72 points 5
CB-2026-06-02-1W Weekly 2026-06-01 to 2026-06-08 -2.55% Large fall -4.29 points 5
CB-2026-06-01-1W Weekly 2026-05-29 to 2026-06-05 -2.50% Large fall -3.44 points 5
CB-2026-05-29-1W Weekly 2026-05-29 to 2026-06-05 -2.50% Large fall -2.86 points 5
CB-2026-05-28-1W Weekly 2026-05-28 to 2026-06-04 +0.33% Little change +2.77 points 5
CB-2026-05-27-1W Weekly 2026-05-26 to 2026-06-02 +1.20% Small rise +3.98 points 4
CB-2026-05-24-1W Weekly 2026-05-22 to 2026-05-29 +1.45% Small rise +2.22 points 4