Anthropic's Claude Opus 4.8 Arrives 41 Days Later With 4x Fewer Silent Bugs
Anthropic's Claude Opus 4.8 lands in Factory's Droid platform with 3x cheaper fast mode, better honesty, and a new Dynamic Workflows feature for codebase-scale agent tasks
- Factory AI's Droid now runs Opus 4.8: Anthropic released Claude Opus 4.8, and Factory AI immediately integrated it into their Droid platform.
- Honesty as a headline feature: Opus 4.8 is ~4x less likely than 4.7 to let flaws in its own code pass unremarked, with alignment scores near the restricted Mythos Preview.
- Fast mode is 3x cheaper: Fast mode now costs $10/$50 per million tokens and runs at 2.5x speed -- three times cheaper than fast mode on previous models.
- Dynamic Workflows ships in research preview: Claude Code can now spawn hundreds of parallel subagents in a single session for codebase-scale migrations (Enterprise, Team, Max plans).
- Same base pricing, new API capabilities: Regular pricing unchanged at $5/$25 per million tokens; mid-conversation system messages and lower prompt cache minimums now supported.
- Mythos coming soon: Anthropic signaled Mythos-class models will be available to all customers "in the coming weeks," making Opus 4.8 likely the near-term mid-tier.
Anthropic has released Claude Opus 4.8, and Factory AI was quick to bring it to Droid, its AI software engineering platform. The upgrade lands with a notably honest self-assessment from Anthropic: the company describes it as "a modest but tangible improvement on its predecessor" -- a refreshingly candid framing for a frontier model launch. But modest does not mean inconsequential. The release bundles meaningful gains in agentic reliability, a dramatically cheaper fast mode, and a new orchestration primitive for large-scale coding tasks that changes what a single agent session can accomplish.
What actually changed
Agentic coding score increases from 64.3% to 69.2%, and multidisciplinary reasoning with tools jumps from 54.7% to 57.9%. On SWE-bench Verified it scores 88.6% versus 87.6% for Opus 4.7, and 69.2% on the harder SWE-bench Pro versus 64.3%. These are incremental gains, but the more interesting story is behavioral rather than benchmark-driven.
The headline improvement is honesty. A general problem with AI models is that they sometimes jump to conclusions, confidently claiming to have made progress despite thin evidence. Early testers report that Opus 4.8 is more likely to flag uncertainties and less likely to make unsupported claims -- borne out in evaluations showing it is around four times less likely than its predecessor to allow flaws in code it has written to pass unremarked.
That behavioral shift matters enormously in agentic settings. An agent that silently ships broken code is far more dangerous than one that stops and says it is not sure. In Claude Code, it asks the right questions, catches its own mistakes, pushes back when a plan is not sound, and builds up confidence around complex, multi-service explorations before making big changes.
The alignment angle
Anthropic's alignment team concluded that Opus 4.8 "reaches new highs on measures of prosocial traits like supporting user autonomy and acting in the user's best interest," and that rates of misaligned behavior such as deception or cooperation with misuse are substantially lower than Opus 4.7, and similar to their best-aligned model, Claude Mythos Preview. That last comparison is significant: Mythos is Anthropic's restricted frontier model currently available only to a handful of organizations for cybersecurity work. Getting Opus 4.8's alignment scores close to Mythos-level while keeping it generally available is a meaningful safety milestone.
Dynamic Workflows: agents at scale
The most architecturally interesting addition is not the model itself but a new feature shipping alongside it. Dynamic Workflows is a Claude Code research preview that lets a single orchestrator session spawn hundreds of parallel subagents, each with its own context window, then aggregate results into a single coherent output. Think of it as the orchestrator-workers pattern -- where one agent breaks a big task into subtasks and delegates them to specialized workers -- baked directly into Claude Code as a first-class primitive rather than something you have to build yourself.
Claude Code alongside Opus 4.8 can now carry out codebase-scale migrations across hundreds of thousands of lines of code from kickoff to merge, with the existing test suite as its bar. Dynamic Workflows is available on Claude Code's Enterprise, Team, and Max plans.
Effort control and the new API behavior
Two quieter but developer-relevant changes shipped alongside the model:
- Effort control: Users on claude.ai can now adjust the computational effort Claude expends on a task, balancing response speed with depth. On Claude Opus 4.8, the effort parameter defaults to high on all surfaces, including the Claude API and Claude Code. You can push it to
xhighfor demanding async jobs, or dial it down to save tokens on simpler tasks. - Mid-conversation system messages: Claude Opus 4.8 accepts
role: "system"messages immediately after a user turn in the messages array. This lets you append updated instructions later in a long-running conversation without restating the full system prompt, which preserves prompt cache hits on the earlier turns and reduces input cost on agentic loops. - Lower prompt cache minimum: The minimum cacheable prompt length on Claude Opus 4.8 is 1,024 tokens, lower than on Claude Opus 4.7. Prompts that were too short to cache on Opus 4.7 can now create cache entries with no code changes.
The mid-conversation system message change is easy to overlook but practically useful. Long agent runs often need to update permissions or token budgets mid-task. Previously that required a full context restart or a workaround. Now it is a clean API call that does not blow up your cache.
Pricing and fast mode
Claude Opus 4.8 is available at the same price as Opus 4.7 -- $5 per million input tokens and $25 per million output tokens. The bigger news is fast mode: fast mode costs $10 per million input and $50 per million output. Anthropic says fast mode is roughly 2.5x faster and three times cheaper than fast mode was for previous models. That price cut makes latency-sensitive agent workloads significantly more viable at scale.
For Factory's Droid users specifically, the tweet highlights that fast mode is now 3x less expensive than before -- a direct cost improvement for teams running high-frequency agent loops where speed matters as much as quality.
Where it wins and where it does not
- On Anthropic's Super-Agent benchmark, 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.
- It delivers the highest score recorded on the Legal Agent Benchmark, and is the first model to break 10% overall on the all-pass standard.
- It scores 84% on Online-Mind2Web for computer use and browser agents, a meaningful jump over both Opus 4.7 and GPT-5.5.
- GPT-5.5 wins on terminal and CLI workflows and is roughly tied on web browsing and graduate-level science.
The bigger picture
The new model comes just 41 days after Opus 4.7 was released, a much faster upgrade cycle than normal for Anthropic. The fast turnaround may have something to do with the chilly reception to Opus 4.7, which some users found disappointing. That interval also saw significant new releases for OpenAI's Codex and Google's Gemini Flash model, increasing the pressure on Anthropic to keep pace.
The release also serves as the clearest signal yet about what comes next. Anthropic wrote: "We're making swift progress on developing these safeguards and expect to be able to bring Mythos-class models to all our customers in the coming weeks." Mythos is currently restricted to a small number of organizations under Project Glasswing for cybersecurity scanning. When it opens up, Opus 4.8 will likely become the mid-tier option -- which means the current flagship pricing may look very different in a matter of weeks.
For teams already on Claude, the upgrade path is straightforward: the API model ID is claude-opus-4-8. It is available immediately on the Claude API, Amazon Bedrock, Google Cloud's Vertex AI, and GitHub Copilot. No pricing changes to negotiate, no prompt rewrites required for most workloads -- just a more honest, more capable agent waiting on the other end.