A dynamic comparison of each model's allocation style across official frozen portfolios. The report separates risk appetite, concentration, defensive ballast, peer overlap, turnover, and resolved performance profile in one view.
Summaries are generated from deterministic metrics. NVIDIA-assisted wording can polish future summaries, but the metric rows below remain the source of truth.
Early allocations lean toward Semiconductors (SMH), Biotechnology (XBI), Taiwan Equities (EWT). The behavioral read should stay provisional until more official portfolios resolve.
Only 2 official saved portfolios so far. Treat the behavior label as provisional.More institutional allocation style. It keeps broader portfolios, carries 1.1% defensive exposure on average, and changes positions less than more tactical peers.
Early allocations lean toward Semiconductors (SMH), Biotechnology (XBI), Financials Sector (XLF). The behavioral read should stay provisional until more official portfolios resolve.
Only 6 official saved portfolios so far. Treat the behavior label as provisional.Concentrated high-conviction style. It uses fewer holdings than diversified peers and gives more weight to the largest position in each portfolio.
Most concentrated structure. It averages 3.56 holdings and a 39.4% largest position, so its portfolios express fewer, higher-conviction views.
Balanced allocator profile. The current data shows 73.4 / 100 average risk-taking, 7.4% defensive exposure, and 27.4% average largest holding.
More institutional allocation style. It keeps broader portfolios, carries 16.6% defensive exposure on average, and changes positions less than more tactical peers.
Balanced allocator profile. The current data shows 72.1 / 100 average risk-taking, 11.9% defensive exposure, and 29.2% average largest holding.
These are cumulative model-level measures across official saved portfolios and resolved results. A lower average rank is better. Beat S&P shows resolved rounds where model return exceeded the S&P 500.
| Model | Risk | Holdings | Top holding | High risk | Defensive | Peer overlap | Turnover | Avg rank | 1st / last | Beat S&P |
|---|---|---|---|---|---|---|---|---|---|---|
| GPT-5.6 Sol | 90.4 / 100 | 5.00 | 42.5% | 100.0% | 0.0% | 79.7% | n/a | n/a | 0 / 0 | 0/0 |
| GPT-5.5 | 85.7 / 100 | 4.89 | 36.0% | 92.0% | 1.1% | 55.9% | 43.8% | 3.64 | 5 / 14 | 12/36 |
| Grok 4.5 | 84.5 / 100 | 4.83 | 33.3% | 93.3% | 0.0% | 66.3% | 65.0% | n/a | 0 / 0 | 0/0 |
| Grok 4.3 | 79.3 / 100 | 3.77 | 37.5% | 78.4% | 4.6% | 57.8% | 53.0% | 2.58 | 11 / 3 | 20/36 |
| Gemini 3.1 Pro | 76.0 / 100 | 3.56 | 39.4% | 69.9% | 9.2% | 50.2% | 57.9% | 3.56 | 3 / 10 | 11/36 |
| Claude Fable 5 | 73.4 / 100 | 5.00 | 27.4% | 65.5% | 7.4% | 60.0% | 56.2% | 3.17 | 1 / 1 | 3/6 |
| Claude Opus 4.7 | 72.5 / 100 | 4.93 | 31.5% | 63.9% | 16.6% | 61.3% | 48.7% | 2.92 | 5 / 5 | 18/36 |
| Claude Opus 4.8 | 72.1 / 100 | 4.98 | 29.2% | 57.6% | 11.9% | 59.2% | 46.0% | 2.42 | 11 / 3 | 14/31 |
Each finding is tied to model IDs and metric keys in the generated report.
GPT-5.6 Sol stands out by risk appetite at 90.4 / 100, while Gemini 3.1 Pro stands out by portfolio structure with a 39.4% average largest holding.
Claude Opus 4.7 has the highest defensive allocation at 16.6%. GPT-5.5 has the lowest measured turnover at 43.8%.
GPT-5.5 has 5 first-place and 14 last-place finishes. Grok 4.3 has 11 first-place and 3 last-place finishes. Gemini 3.1 Pro has 3 first-place and 10 last-place finishes. Claude Opus 4.7 has 5 first-place and 5 last-place finishes. Claude Opus 4.8 has 11 first-place and 3 last-place finishes.
GPT-5.6 Sol has the highest average peer overlap at 79.7%. This means its allocation weights have looked more like the rest of the roster than the most distinctive models.
The report is generated from official frozen portfolios and resolved result rows during the website build. No page copy is manually assigned to a model. New models appear automatically after they have official saved portfolios.
Models with fewer than 8 saved portfolios are marked as early sample. Performance language is caveated until at least 3 resolved results are available.
NVIDIA-assisted text is allowed only as a rewrite layer. The prompt receives the structured model rows, traits, metric keys, top assets, and comparative candidates. It is not allowed to add unsupported numbers, assets, causes, stale dates, or investment advice.
Prompt contract: capitalbench_model_patterns_prompt_v1
Average allocation-weighted risk appetite across all official saved portfolios. Higher means more growth, momentum, cyclical, and high-risk exposure.
Average number of non-zero assets in the model's official saved portfolios.
Average size of the largest single holding in each official saved portfolio.
Average allocation to assets rated as higher risk by the CapitalBench asset risk model.
Average allocation to cash, bonds, defensive sectors, and other lower-risk ballast.
Average allocation to technology, semiconductors, Nasdaq-style growth, and AI-linked technology exposure.
Average allocation to cash-like assets and duration-sensitive bond exposure.
Average allocation to non-U.S. country, regional, or international equity exposure.
Average allocation to commodities, crypto, energy, gold, and other inflation-linked or real-asset groups.
Average cosine similarity between this model's allocation weights and peer model portfolios in the same rounds.
Average one-half summed absolute allocation change between consecutive same-track portfolios.
Average finishing rank across resolved rounds. Lower is better.
Average model score versus the hindsight-best eligible asset in each resolved round.
