Opus 4.8 × Oxygen

Claude Opus 4.8 is out. Here is why it's big for GTM.

GTM is tool-calling work. Pull a list, enrich it, check the CRM, update a field, send the sequence, read the reply. The model is not writing prose. It is running tools against your stack, in order, without dropping a step.

So the agentic benchmarks matter more than the chat ones.

Opus 4.8 tops the agentic suite: - SWE-Bench Pro (agentic coding): 69.2% vs GPT-5.5 at 58.6% - OSWorld (computer use): 83.4% vs GPT-5.5 at 78.7% - Finance Agent v2: 53.9%, top of the field

(GPT-5.5 still edges terminal coding, 78.2% vs 74.6%. Credit where it's due.)

More reliable tool calls means fewer broken runs.

The underrated part is honesty.

Opus 4.8 is about 4x less likely to let a flaw in its own work pass unremarked than 4.7. It flags what it is unsure about. It makes fewer unsupported claims.

In a chat window that is a nice-to-have. In your CRM it is everything.

An agent that invents a job title or guesses company size because it sounded right is worse than no agent. The honesty gains are what keep it grounded in your data instead of filling gaps with plausible BS.

That is the model. Wiring it to your data, tools, and workflows is still on you.

If you want to run Opus 4.8 across your GTM stack, that is what we built OXYGEN for. MCP or CLI, install in minutes.

Release notes in the comments.