What did Anthropic test?
Anthropic says it tested Claude models on trading in EVE Online without placing trades in the live game. At Fenris Creations’ request, the company kept automated trading out of live EVE Online, where Anthropic says it is against the rules. Instead, Anthropic built a simulator that replayed real buy and sell orders posted by players in the game’s largest and most competitive trading hub, while applying fees and trading restrictions faced by human traders.
How were the models evaluated?
In the trading tests, Anthropic says Claude Opus 4.6, Opus 4.7, Opus 4.8 and Fable 5 each received 3 billion ISK, the game’s currency, and instructions to maximize profit by the end of the trading period. The models could also read in-game news and patch notes as they came out. Anthropic says the models were not told how to trade. Most independently adopted spread capture: buying from immediate sellers, reselling to immediate buyers and keeping the difference after fees.
What results did Anthropic report?
Anthropic reports that, across nine two-week simulated trading windows, Opus 4.7, Opus 4.8 and Fable 5 were profitable in every window. Their average capital growth was about 16%, 22% and 43%, respectively, while Opus 4.6 gained about 4%. In the main experiments, which lasted roughly six months and covered October 2024–April 2025 and October 2025–March 2026, average earnings rose from around 200% of initial capital for Opus 4.6 to 600% for Fable 5.
Analysis: What the setup can and cannot show
The reported design matters because it separates two questions. The simulator used real player orders, fees and trading restrictions, so the results describe how the models performed against a replayed market record under Anthropic’s test conditions. But because Anthropic says it did not trade in live EVE Online, the figures are not evidence that Claude traded with current players or that automated trading was deployed in the live game. The two-week and roughly six-month results also measure different horizons, so the shorter-window profitability and longer-run earnings should not be treated as the same statistic.
Official announcement · Additional coverage
Official announcement published: November 06, 2026 at 00:00 UTC.
