Anthropic Releases Claude Opus 5, Claims Near-Fable-5 Performance at Opus Pricing

Anthropic released Claude Opus 5 on July 24, 2026, positioning it as a coding- and enterprise-focused upgrade that the company says delivers performance close to its flagship Fable 5 model at a fraction of the cost. Pricing remains unchanged from Opus 4.8: $5 per million input tokens and $25 per million output tokens via the API.

The release follows a rapid cadence of Anthropic model launches — Mythos 5 and Fable 5 last month, Sonnet 5 on July 1 — making Opus 5 the fourth major model update in roughly a month.

Performance claims

Anthropic states that "Opus 5 comes close to the capabilities of Claude Fable 5 in many domains at half the price," though the company has not published specific benchmark scores to substantiate this comparison. Cursor co-founder Sualeh Asif said, according to the company, that on CursorBench Opus 5 scores "just under Fable 5" and exhibits "many of the same behaviors" — again without disclosed numeric results.

Harvey, a legal AI vendor, reported that Opus 5 achieved similar output quality while generating 26% fewer tokens on average compared to Opus 4.8 at maximum reasoning settings, according to Niko Grupen, the company's head of applied research. Anthropic frames this token efficiency as a proxy for reduced computational strain during complex, multi-step tasks, though token count is not a standard accuracy benchmark.

Anthropic also claims Opus 5 "overperforms on biology tasks compared to Opus 4.8," without defining the metric or providing comparative data.

What's new

Beyond raw model capability, Anthropic is shipping three related features:

• Fast mode (research preview): Full Opus 5 capability at a claimed 150% faster token generation, priced at double the standard rate. Claude Code Max subscribers can access it via additional usage credits.
• Automatic fallbacks: When a safety classifier declines a request, the API now automatically retries with a different model within the same request rather than returning an error, according to Anthropic.
• Mid-conversation tool changes: Developers can add or remove external tools mid-conversation without invalidating the prompt cache — intended to cut costs and reduce the attack surface by disabling unused tools.

Opus 5 is available immediately to all account tiers that previously had Opus 4.8 access, including standard usage limits on paid plans. Anthropic continues to recommend its separate Fable 5 model for what it calls "days-long autonomous projects," suggesting Opus 5 is positioned as the default workhorse rather than a replacement for Fable 5 in the most demanding agentic workloads.

Context window and detailed benchmark figures for Opus 5 were not disclosed in Anthropic's announcement; Opus 4.8 supported a 1-million-token context window.

What this means

Anthropic is betting that token efficiency and workflow features — not just raw benchmark wins — will drive adoption of Opus 5, particularly among enterprise and coding customers already anchored to Opus pricing. The lack of published benchmark scores for the Fable 5 comparison makes the "near-Fable performance" claim difficult to verify independently; buyers evaluating the model for production use should request task-specific evaluations rather than relying on vendor and partner testimonials alone. The mid-conversation tool-switching feature addresses a real pain point in agentic API design — previously, changing available tools mid-session broke prompt caching and increased cost — and is likely to see fast adoption among developers running long agent sessions. Anthropic's decision to hold pricing steady while claiming capability gains suggests the company is prioritizing retention of existing API customers over introducing new price tiers, at least for this release.