Claude Opus 4.8 is Live in Kilo Code—Grab It Now at 20% Off!
The new edition of this frontier model is more reliable, more efficient, and currently available at a discount in Kilo
Anthropic’s latest mega release, Claude Opus 4.8, is live in Kilo Code. To celebrate, we’re offering a flat 20% discount on Opus 4.8 inference through our provider network.
If you’ve been following recent posts about affordable tokenmaxxing and BYOK setups (with providers like xAI, Fireworks AI and Anthropic), you know we love passing savings directly to you. This latest release combined with an exclusive discount is our way of saying thanks to the community.
Opus 4.8 is a massive leap forward for code-gen and autonomous workflows. But, let’s be clear, there have been a lot of breakthroughs recently with both OSS and private models. The community is scaling—not just one lab or one model.
Anthropic’s particular focus with this release was on honesty. That can be a nebulous term when it comes to LLMs, but the lab has a nice tight definition for what they were trying to accomplish here:
We train all our models to be honest—for instance, to avoid making claims that they can’t support. But a general problem with AI models is that they sometimes jump to conclusions, confidently claiming to have made progress in their work despite the evidence being thin. Early testers report that Opus 4.8 is more likely to flag uncertainties about its work and less likely to make unsupported claims. This is borne out in our evaluations, which show that Opus 4.8 is around four times less likely than its predecessor to allow flaws in code it has written to pass unremarked.
Our initial tests have found this to be true, similar to the improvements in rule following with the release of OpenAI’s GPT-5.5. The confidence that comes with higher intelligence is being tempered—or subject to more restraints—with these new frontier releases.
So how does the new Opus stack up overall?
The Benchmarks: Crushing the Competition
When it comes to dev capabilities, Opus 4.8 brings serious heat to the table.
PinchBench: Opus 4.8 absolutely crushed PinchBench, scoring an impressive average of 90.5% and succeeding across categories and tasks for both generation and analysis.
Fast Mode: Running Opus 4.8 in “Fast Mode” (which processes at 2.5x the speed) pushes that benchmark score even higher. Opus 4.7 Fast was also really good but also perilously expensive; fast mode for 4.8 is one third the cost.
KiloBench: On our internal KiloBench eval, standard Opus 4.8 performed similarly in intelligence to Opus 4.7, but ran for 15% less cost ($85.19 vs $100.51). Opus 4.8 appears to be less expensive for similar tasks.
Agentic Reliability: Anthropic reports that Opus 4.8 is 4x less likely to let flaws in code it has written pass unremarked. It noticeably fixes the verbosity and tool-calling bottlenecks some developers saw with Opus 4.7.
So nobody would call it cheap, but improvements in efficiency and a lower base price on fast mode are making Opus 4.8 a strong contender for enterprise users and well-funded startups.
What’s New for Agent Builders?
With Opus 4.8, Anthropic didn’t just bump the model weights. They shipped vital orchestration features purpose-built for agentic engineering.
Some of the stuff we’re most excited about in this new release:
Dynamic Workflows: Opus 4.8 is built to tackle codebase-scale migrations. It can plan a massive task, spin up hundreds of parallel subagents in a single session, and verify outputs against your existing test suite before reporting back.
Mid-Task System Prompts: The Messages API now accepts system entries inside the messages array. This means you can update token budgets, agent permissions, or environment context mid-flight without breaking your prompt cache.
Granular Effort Control: You can now throttle how much compute Claude spends on a given task. Dial it up to “extra” or “max” for high-stakes, asynchronous multi-service debugging, or dial it down to save on rate limits.
Computer-Use Dominance: Opus 4.8 is currently the strongest computer-use model Anthropic has tested, scoring an 84% on Online-Mind2Web for highly reliable end-to-end agent workloads. This is great for KiloClaw!
How will you use Opus 4.8 in your Kilo Code workflows? The possibilities are endless.
Get Building
Claude Opus 4.8 is available right now in your Kilo Code environment. Just toggle the model dropdown, choose whether you want to route through our discounted stealth provider for 20% off or use the direct API connection for strict data privacy, and let your agents loose.
To get the 20% discount, just make sure to select the correct model in the model selector dropdown. The discounted models appear as separate listings.
Happy coding!
Check out the full Anthropic Opus 4.8 Release Notes and the model page on Kilo for more details about the new model.