Anthropic's Claude Opus 4.8 Hits 69.2% on SWE-Bench and Lands in Devin
Claude Opus 4.8 lands in Windsurf and Devin CLI with better agentic judgment, 3x cheaper fast mode, and dynamic parallel subagent workflows
Takeaways
- Claude Opus 4.8 is now live in Windsurf (Devin Desktop) and the Devin CLI for all paid subscribers.
- Key improvement: Opus 4.8 is ~4x less likely than Opus 4.7 to silently pass over flaws in code it writes — critical for unattended agent runs.
- Dynamic Workflows (research preview) lets Claude Code spawn hundreds of parallel subagents in a single session for codebase-scale tasks.
- Fast Mode is 3x cheaper than on prior Opus models, while still delivering 2.5x output speed.
- Benchmarks: Leads on SWE-Bench Pro (69.2%), Online-Mind2Web (84%), and is the only model to complete every Super-Agent benchmark case end-to-end.
- API pricing unchanged at $5/$25 per million tokens; access via claude-opus-4-8.
Anthropic's Claude Opus 4.8 is now available inside Windsurf (now rebranding to Devin Desktop) and the Devin CLI. It's the most capable model yet to land in the IDE, and it arrives with a meaningful set of improvements that matter specifically for the kind of long-running, agentic coding sessions these tools are built around.
What changed under the hood
Opus 4.8 is an upgrade to Anthropic's Opus class of models, with stronger performance across coding, agentic tasks, and professional work, and the consistency to handle long-running work. The headline number: Opus 4.8 leads in 6 out of 7 benchmarks, achieving the highest score on SWE-Bench Pro at 69.2%.
But the more interesting improvement is behavioral. Early testers found Opus 4.8 to be more reliable and sharper in its judgment when performing agentic tasks. Specifically, the model is significantly less likely to silently paper over its own mistakes. According to Anthropic's evaluations, Opus 4.8 is around four times less likely than its predecessor to let flaws in code it has written pass unremarked , a critical property when you're trusting an agent to run unattended.
Opus 4.8 uses tools cleanly and follows instructions with the consistency autonomous engineering workloads need to keep running unattended. It improves on Opus 4.6 and fixes the comment-verbosity and tool-calling issues seen with Opus 4.7. Cognition's CEO Scott Wu noted directly that this release from Anthropic translates directly into faster capability gains for engineers building on Devin.
What's new for Windsurf and Devin CLI users
Beyond the base model upgrade, Opus 4.8 ships alongside a few features worth knowing about:
- Dynamic Workflows (research preview): This new feature allows Claude to take on even bigger tasks in Claude Code. Claude can plan the work and then run hundreds of parallel subagents in a single session, with Opus 4.8 agents able to run for even longer. Think codebase-scale migrations across hundreds of thousands of lines of code, kicked off in one session and carried through to merge.
- Fast Mode, now 3x cheaper: Fast mode for Opus 4.8 , where the model works at 2.5x the speed , is now three times cheaper than it was for previous models. This makes the speed/cost tradeoff much more practical for interactive workflows.
- Effort Control: Users on claude.ai now have control over the amount of effort Claude puts into a task. In Claude Code, the default is
higheffort, withxhighandmaxavailable for harder tasks and async workflows. - Mid-conversation system messages: The Messages API now accepts system entries inside the messages array, letting developers update Claude's instructions mid-task without breaking the prompt cache.
Benchmark highlights
On Anthropic's Super-Agent benchmark, Claude Opus 4.8 is the only model to complete every case end-to-end, beating prior Opus models and GPT-5.5 at parity on cost. On the browser-agent side, Opus 4.8 is the strongest computer-use and browser-agent model tested, scoring 84% on Online-Mind2Web , a meaningful jump over both Opus 4.7 and GPT-5.5.
On CursorBench, Claude Opus 4.8 exceeds prior Opus models across every effort level. Tool calling is meaningfully more efficient, using fewer steps for the same intelligence, and it carries end-to-end tasks through. The one area where Opus 4.8 doesn't lead: GPT-5.5 remains the leader in Terminal-Bench 2.1 at 78.2%, indicating stronger terminal-based agent performance.
Pricing and availability
Opus 4.5 is available to all paid Windsurf subscribers. For API access, pricing for Opus 4.8 starts at $5 per million input tokens and $25 per million output tokens, with up to 90% cost savings with prompt caching and 50% savings with batch processing. Fast mode pricing runs $10 per million input tokens and $50 per million output tokens , but remember, that's 3x cheaper than fast mode was on Opus 4.7. Developers can access it via claude-opus-4-8 through the Claude API.
The broader context here is that Windsurf itself is in the middle of a significant transition. Cognition has relaunched Windsurf as Devin Desktop, a full IDE with an Agent Command Center and open ACP protocol support that lets Codex, Claude Agent, OpenCode, and custom in-house agents run side by side in a shared Kanban workspace. Opus 4.8 landing in both the IDE and the Devin CLI simultaneously signals that the platform is positioning itself as a serious multi-model, multi-agent orchestration layer , not just a single-model coding assistant.
For teams running long autonomous coding sessions, the combination of Opus 4.8's improved honesty, cleaner tool use, and the new dynamic workflows feature is the most practical upgrade here. The model is less likely to confidently hallucinate progress, and when it does hit a problem, it's more likely to flag it rather than silently move on.