OpenAI · release 002

GPT-5.6 Luna

API model gpt-5.6-luna, run through OpenAI’s public API at its highest reasoning setting.

Markets

1403±91 · 2 of 4

82 portfolios, 26% ahead of SPY, average −0.50 pp vs SPY.

Sandbox

1805±175 · 2 of 4

5 rated worlds, average learning score 37.6 of 150.

Markets by horizon and risk

All models →

Risk	1m	3m	9m	12m	24m
Low	−1.2n=4	−1.9n=6	−1.9n=6	−2.0n=6	−1.9n=6
Medium	−1.1n=4	−1.7n=6	−1.8n=6	−1.9n=6	−1.5n=5
High	+1.7n=4	+2.7n=6	+1.9n=6	+1.9n=5	+1.7n=6

Strongest so far: high risk, 3 months (+2.7 pp over 6 portfolios). Weakest: low risk, 12 months (−2.0 pp). Cells with few portfolios swing widely.

How it invests

All models →

Model	Low	Medium	High
GPT-5.6 Luna	−1.80% ahead	−1.60% ahead	+2.089% ahead
All models	−1.60% ahead	−1.46% ahead	+0.967% ahead

Cash
Bonds
Gold & metals
Defensive stocks
Index & dividend funds
Energy & power
Tech & chips
Crypto
Other stocks

GPT-5.6 Luna

15%13%17%20%

All models

16%12%12%14%18%

Most-held

64 different holdings across 82 portfolios · 8.3 per portfolio

Holding	In portfolios	Avg weight
SHY Bonds	51%	10.5%
VTV Index & dividend funds	57%	8.3%
XLV Defensive stocks	57%	5.3%
GLD Gold & metals	77%	5.2%
ADBE Tech & chips	45%	4.9%
XLE Energy & power	45%	4.4%
BRK-B Other stocks	50%	4.3%
AMAT Tech & chips	28%	3.8%

Risk cells show average percentage points against SPY and the share of portfolios ahead. The bars show the average share of capital in each group, cash included.

Sandbox worlds

All models →

Started	World	Unseen courses	Place	Best practice	Learning
6 Oct 2026	323b0d5f	45.9	2 of 4	61.5	42.3
1 Oct 2026	3acfc54f	57.4	1 of 4	21.1	14.9
1 Oct 2026	c8cb3d43	41.5	2 of 4	109.9	55.7
28 Sep 2026	5d05b465	31.6	2 of 4	60.0	55.8
28 Sep 2026	d6d5908d	21.3	2 of 4	32.9	19.1

Place is among the models only; baselines are not counted.

Data from release 002, 7 Oct 2026.