Claude Opus 5 matches Fable 5 performance at half the cost

Key Takeaways

• Claude Opus 5 achieves near-Fable 5 performance while costing half as much at $5/$25 per million input/output tokens
• The model uses 26% fewer tokens on average compared to Opus 4.8 at maximum reasoning
• Anthropic positions Opus 5 as the daily driver for enterprise knowledge work and coding

Anthropic released Claude Opus 5 today, completing its model lineup refresh that began with Mythos 5 and Fable 5 last month and continued with Sonnet 5 on July 1. The upgrade brings near-Fable 5 intelligence to the Opus tier at half the price, with Anthropic claiming substantial improvements in coding, knowledge work, and token efficiency.

The pitch is straightforward: get Fable-level performance without paying Fable prices. API pricing stays unchanged from Opus 4.8 at $5 per million input tokens and $25 per million output tokens. For teams running heavy workloads through Claude, the cost-to-capability ratio matters more than raw benchmark scores.

How does Opus 5 compare to Fable 5?

Anthropic's own framing is careful: Opus 5 "comes close to" Fable 5 in many domains, not all. Cursor, the AI coding tool, tested this claim directly. Co-founder Sualeh Asif says Opus 5 scores "just under Fable 5" on CursorBench and exhibits many of the same behaviors. That's a meaningful benchmark because Cursor users push models hard on real code.

The tradeoff? Speed and cost. Opus 5 runs faster and cheaper than Fable 5, making it viable as an always-on assistant rather than a model you call sparingly for complex tasks.

26% fewer tokens for the same work

Harvey, the AI legal tools company, ran Opus 5 against Opus 4.8 on legal agent work. According to Niko Grupen, Harvey's head of applied research, Opus 5 maintained similar performance while generating 26% fewer tokens on average at maximum reasoning.

Token efficiency isn't just about cost. Fewer tokens mean the model reaches conclusions faster and exhausts context windows more slowly. For long agentic runs, where Claude Code often requests breaks to "nap," using fewer tokens per task extends how much work the model can complete in a single session.

What's improved for developers?

Stripe software engineer Cristian Rivera describes Opus 5 as "a clear generational step up from Opus 4.8." Over one weekend, he gave the model a chief-of-staff role over his development environments. It built its own monitor, drove each box, and pulled him in only for judgment calls.

Anthropic lists several developer-facing improvements: greater autonomy with less back-and-forth, better error recovery, and more frequent self-checks. The practical effect is fewer interruptions during complex tasks. Combined with prompt-cache-friendly tool changes, teams running automated pipelines should see smoother execution.

The company also highlights biology and scientific research gains, claiming Opus 5 "overperforms" on biology tasks compared to Opus 4.8. Anthropic hasn't defined that metric, but the emphasis signals continued investment in making Claude useful for specialized domains beyond software.

Who should switch to Opus 5?

Anthropic positions Opus 5 as the enterprise daily driver. For Claude Max subscribers paying $100/month, the switch should be automatic. One heavy user reported running large agentic workloads all week and hitting only 31% of the usage limit. With Opus 5's efficiency gains, that runway extends further.

API users paying per token face a simpler calculation. If Opus 4.8 handled your workload, Opus 5 does the same work faster and with fewer tokens. If you needed Fable 5 for quality, test whether Opus 5 hits your accuracy threshold. At half the cost, even a small quality gap might be worth accepting.

The model tier gap is narrowing

Anthropic's release cadence tells a story. Fable 5 launched with frontier capabilities and frontier pricing. Weeks later, Opus 5 arrives with "near-Fable" performance at existing Opus rates. The gap between tiers shrinks with each release.

For buyers, this creates a waiting game. Pay premium prices for immediate access, or wait a few weeks for similar capabilities at lower cost. Anthropic likely structures this intentionally, capturing maximum revenue from users who need cutting-edge performance immediately while expanding the addressable market as costs drop.

Frequently Asked Questions

How much does Claude Opus 5 cost?

Claude Opus 5 costs $5 per million input tokens and $25 per million output tokens via API, the same as Opus 4.8. Claude Max subscribers at $100/month get Opus 5 included.

Is Claude Opus 5 better than Claude Fable 5?

Opus 5 comes close to Fable 5 capabilities in many domains but scores slightly below it on benchmarks like CursorBench. The tradeoff is cost: Opus 5 runs at half the price.

How much more efficient is Claude Opus 5 than Opus 4.8?

According to Harvey's testing, Opus 5 generates 26% fewer tokens on average compared to Opus 4.8 at maximum reasoning while maintaining similar quality.

When was Claude Opus 5 released?

Anthropic released Claude Opus 5 on July 24, 2026, following the releases of Mythos 5, Fable 5, and Sonnet 5 in the preceding weeks.

What is Claude Opus 5 best at?

Anthropic highlights coding, knowledge work, and agent workflows as primary strengths. Third-party testers report gains in legal work, software development, and biology research tasks.

Source: Latest news