Claude Opus 4.8 lands in Kilo Code with 20% discount
Claude Opus 4.8 is now available in Kilo Code, and the company is offering a 20% discount on inference through its provider network. Kilo says the model is more reliable and efficient, with new orchestration features for agentic workflows and computer-use tasks.
Claude Opus 4.8 is now live in Kilo Code, and Kilo is offering a flat 20% discount on Opus 4.8 inference through its provider network. The company said users can choose between the discounted provider route or a direct API connection for stricter data privacy.
Kilo described the launch as part of its push around affordable tokenmaxxing and bring-your-own-key, or BYOK, setups. It said the discount is meant to pass savings directly to users and follows what it called a broader wave of progress across both open-source and private models.
According to Kilo, Anthropic focused this release on honesty. The company said Opus 4.8 is designed to avoid unsupported claims and is more likely to flag uncertainty about its work instead of overstating progress.
Kilo said its early tests show the model is around four times less likely than its predecessor to let flaws in code it wrote go unremarked. The company also said those tests lined up with improvements it saw in OpenAI’s GPT-5.5 around rule following.
On the performance side, Kilo said Opus 4.8 posted strong results on its internal and external evaluations. It said the model scored 90.5% on PinchBench, performing well across generation and analysis tasks, and that Fast Mode, which processes at 2.5x the speed, pushed benchmark scores even higher.
Kilo also said standard Opus 4.8 matched Opus 4.7 in intelligence on its internal KiloBench while costing less. The company cited a cost of $85.19 for Opus 4.8 versus $100.51 for Opus 4.7, a drop of about 15%.
For agent builders, Kilo said the release includes workflow features aimed at larger automation jobs. Those include dynamic workflows for codebase-scale migrations, the ability to spin up hundreds of parallel subagents in one session, and verification against an existing test suite before reporting results.
The Messages API now accepts system entries inside the messages array, according to Kilo. The company said that change lets developers update token budgets, agent permissions, or environment context mid-task without breaking the prompt cache.
Kilo also said developers can now control how much compute Claude uses on a task by dialing effort up to “extra” or “max” for heavier debugging jobs, or lowering it to save on rate limits. The company said that should be useful for asynchronous multi-service work.
Anthropic’s computer-use performance is another focal point of the release. Kilo said Opus 4.8 is the strongest computer-use model Anthropic has tested, with an 84% score on Online-Mind2Web for end-to-end agent workloads.
The model is available now in Kilo Code. Users can select Opus 4.8 in the model dropdown, then choose either the discounted stealth provider option or the direct API connection, according to Kilo. The discounted listing appears separately in the model selector, the company said.
