Anthropic launched Claude Opus 5 on July 24, 2026. The model is available immediately on all of Anthropic’s platforms, priced at $5 per million input tokens and $25 per million output tokens: the same cost as its predecessor, Opus 4.8, and half the cost of Fable 5. On Frontier-Bench, ARC-AGI 3, OSWorld 2.0, and Zapier AutomationBench, it sets new state-of-the-art results while outperforming Fable 5 on most benchmarks at a fraction of the price.
For enterprise teams evaluating which model to anchor their AI workflows on, this is the most significant Anthropic pricing signal of 2026: near-frontier capability is now available at mid-tier cost.
What Launched and What It Costs
Claude Opus 5 is available today as claude-opus-5 on the Claude API. It becomes the new default model on Claude Max and the strongest model available on Claude Pro. Pricing is unchanged from Opus 4.8: $5 per million input tokens, $25 per million output tokens. Fable 5, Anthropic’s current top-tier model, carries roughly double those figures.
Three features ship alongside the model:
- Effort setting: an adjustable dial that trades intelligence for speed and token savings. Teams can run at max effort for the hardest tasks and dial down for routine operations.
- Fast mode: approximately 2.5x default speed at twice the base price, for latency-sensitive production workflows.
- Automatic Fallbacks (opt-in API beta): when a prompt triggers Anthropic’s safety classifier, the API routes to a less capable model instead of returning an error. Production workflows that occasionally touch classifier boundaries will get a response instead of a 500.
Consistent with prior Opus models, Opus 5 carries no data retention requirements. This matters for enterprise and regulated-industry teams that have run on Opus 4.8 specifically because of that constraint: the upgrade path is clean.
The Benchmark Picture
The numbers Anthropic published are striking enough to warrant quoting directly, with context:
| Benchmark | Opus 5 | Fable 5 | Opus 4.8 |
|---|---|---|---|
| Frontier-Bench v0.1 | 43.3% | 33.7% | 18.7% |
| ARC-AGI 3 | 3x next-best | below Opus 5 | below Opus 5 |
| OSWorld 2.0 | Beats Fable 5 at 1/3 the cost | reference | below |
| CursorBench 3.2 | Within 0.5% of Fable 5 | reference | below |
| Zapier AutomationBench | 1.5x next-best pass rate | below Opus 5 | below |
Frontier-Bench v0.1 is an agentic terminal coding benchmark. Opus 5 more than doubles Opus 4.8’s score and surpasses Fable 5. On ARC-AGI 3, which tests novel problem-solving, Anthropic says Opus 5 scored three times as high as the next-best model. On OSWorld 2.0, a computer-use benchmark, it exceeds Fable 5’s best result at just over one-third the cost.
Anthropic is also transparent about where Opus 5 falls short. It remains behind Mythos 5 on offensive cybersecurity and advanced biology research. On very long-horizon autonomous tasks, the company says Fable 5 still holds an advantage that benchmarks do not fully capture, because current evals primarily test bounded tasks with defined outcomes, not multi-day work that must sustain coherence across many connected steps. That distinction matters for teams deploying autonomous agents on extended projects.
Early Customer Signals
Three early-access customers provided concrete production numbers:
Harvey (legal AI): Claude Opus 5 achieved similar performance to Opus 4.8’s maximum-reasoning mode while generating 26% fewer tokens on average. For legal workflows where document volume is large, that reduction directly compounds into cost savings.
Fundamental Research Lab: On hard financial modeling tasks, the model averaged nine percentage points higher accuracy while using roughly one-third fewer turns and tool calls and 60% less time. That is a meaningful compounding improvement: better answers, fewer API calls, faster completion.
Zapier: Topped AutomationBench without spending more tokens than prior Claude models, running a full churn-prevention workflow from start to finish. Zapier’s AutomationBench measures whether models can complete business tasks end-to-end without human intervention. Opus 5 passed more tasks than any other model even at its lowest effort setting.
These are real-world workflow numbers across legal, financial, and operations contexts, not synthetic benchmark labs. For GTM and operations teams deciding which model to anchor on, they are more relevant than benchmark tables.
How This Reshapes the Anthropic Model Lineup
Anthropic now offers a clearly tiered lineup, and the guidance is more explicit than it has been:
- Mythos 5: Frontier capability in sensitive domains (cybersecurity, advanced research). Subject to special access controls.
- Fable 5: Maximum autonomy for the longest, most complex projects. Days-long agentic work with dense source material and many connected steps.
- Opus 5: Daily driver. Complex bounded work, coding, knowledge work, agentic automation. Best economics for most enterprise use cases.
- Sonnet 5: High-volume production. Where cost per call and throughput set what ships.
- Haiku 4.5: Subagents, instant responses, the high-frequency layer of multi-agent architectures.
This stratification matters for enterprise architecture decisions. We analyzed how Anthropic’s revenue trajectory has reflected enterprise adoption of Claude’s more capable models in our earlier piece on Anthropic’s enterprise revenue lead. Opus 5 is the product that operationalizes that trajectory: a model capable enough to run real enterprise work, priced to run it at scale without forcing a choice between intelligence and cost.
The Sonnet 5 tier, by contrast, is optimized for scenarios where you want many parallel agents or high API call volumes. We covered the Sonnet 5 launch in June in our analysis of Claude Sonnet 5 as an agentic enterprise model. The full-lineup picture now reads as: Sonnet 5 for scale, Opus 5 for depth, Fable 5 for the frontier edge cases most organizations rarely hit.
The Strategic Signal: Economics of Daily AI Use
Behind the benchmark numbers is a positioning decision worth naming directly. Anthropic is not claiming Opus 5 is the smartest model in any category. What it is claiming, and backing with customer data, is that the most economically important AI work happens in a band of difficulty where near-frontier intelligence delivered efficiently beats frontier intelligence delivered expensively.
That claim, if correct, has significant implications for enterprise AI spending. Organizations that anchored on Fable 5 for everyday coding, document processing, and workflow automation have been paying a frontier-tier premium for bounded tasks that Opus 5 now handles better, at half the cost. The upgrade path is immediate: same API, same pricing, higher throughput on bounded tasks.
The VentureBeat interview with Anthropic’s spokesperson was unusually direct about what separates the two tiers: “Opus 5 is the best tool for the jobs benchmarks can see, and Fable 5 is what you reach for when the job outruns the benchmark.” That framing, bounded tasks versus long-horizon autonomy, is likely to become the primary axis of enterprise model selection through the rest of 2026.
What Enterprise Teams Should Do
| Situation | Recommended action |
|---|---|
| Currently using Opus 4.8 for production workflows | Migrate to claude-opus-5. Same pricing, meaningfully better performance on coding and knowledge work. |
| Using Fable 5 for bounded coding or document tasks | Run Opus 5 on a representative sample. The economics case is strong. |
| Building multi-agent systems with many parallel calls | Sonnet 5 or Haiku 4.5 for the high-frequency layer; Opus 5 for orchestration and complex reasoning steps. |
| Long-horizon autonomous projects spanning multiple days | Fable 5 remains the appropriate anchor for now. |
| Regulated industry with zero data retention requirement | Opus 5 explicitly carries no data retention requirement. Migration path is clean. |
Opus 5 is available today. For teams that have been waiting for a Claude model that delivers near-Fable intelligence at production-friendly economics, the wait is over.
Enera helps enterprise and GTM teams design AI agent architectures and select the right model stack for their workflows. For a focused conversation about how Opus 5 fits your current or planned deployment, book a call with our team.