Claude Opus 5.5 for stock trading

Opus 5.5 is the current Opus, and it costs less than the Opus models before it: $4 per million input tokens and $20 per million output, against $5 and $25 for Opus 4.8. The context window is 1 million tokens and a single response can run to 128K tokens. For a trading agent that changes the math on where Opus belongs. A model that was only affordable for a daily strategy review is now closer to something you can run on a slower trade loop.

The context window matters more than it first looks. A million tokens holds a long history of your own fills, your written reasoning for each, the current portfolio and whatever research you keep, all in one prompt. That makes Opus 5.5 a good fit for the review step: read everything the agent did this week, find the trades where the stated reason and the outcome disagree, and rewrite the rules. Most agents do not need that much context on every pass, so the usual split still holds. Opus reviews and revises, and a cheaper model runs the minute-to-minute decisions.

Thinking is always on. Opus 5.5 does not take a setting that turns reasoning off; the control is the effort level, which defaults to medium. For a trade loop, low or medium effort keeps latency and token spend down. For the weekly review, high effort is worth paying for. If you are moving code from an older Claude model, check two things: requests that turn thinking off, and requests that force a specific tool call, are both rejected on this model.

The honest caution is the same as for any strong model in a trading loop. Give it room to reason over an ambiguous setup and it can build a convincing case for a trade that should not happen. Position limits, entry rules and kill switches belong in code, where no amount of reasoning can talk past them.