Claude Opus 5 Launches: Same Price as Opus 4.8, Within 0.5% of Fable 5 on Coding Benchmarks
Anthropic's most aligned model yet also handles biology tasks Fable 5 blocks - at no price increase.
Claude Opus 5 launched today on all Anthropic platforms, priced at $5 per million input tokens and $25 per million output tokens - identical to what Opus 4.8 costs. On CursorBench 3.2, Opus 5 lands within 0.5% of Fable 5's peak score at max effort while cutting the cost per task in half. Anthropic made it Claude Max's new default model and the strongest option available on Claude Pro.
Frontier-Bench and OSWorld Numbers Back the Headline
On Frontier-Bench v0.1, Opus 5 outscores every other model and more than doubles Opus 4.8's result at lower cost per task. OSWorld 2.0 is the sharper example: Opus 5 beats Fable 5's best computer-use result at just over a third of the cost, and at minimum effort still passes more tasks than any other model. On ARC-AGI 3, a benchmark for genuinely novel problem-solving that models cannot memorise, Opus 5 scores three times higher than the next-best competitor.
Zapier's AutomationBench adds business context. Opus 5's pass rate runs about 1.5 times the next competitor's for identical cost, and even at its lowest effort setting, Opus 5 outperforms every prior model on the leaderboard. For developers comparing token costs across providers, AI API Pricing Compared 2026 has the full per-token breakdown alongside GPT-5.6 and Gemini.
Alignment Score Hits 2.3 - the Lowest in the Claude Family
Anthropic's automated behavioral audit rated Claude Opus 5 at 2.3 on overall misaligned behavior - the lowest among recent Claude models, and below Opus 4.8, Sonnet 5, and Fable 5. On cybersecurity specifically, Opus 5 approaches Mythos 5 at finding software vulnerabilities but falls well behind on developing exploits. Anthropic says that gap is deliberate: the company avoided training Opus 5 on offensive cyber tasks. For context on how Sonnet 5 and Opus 4.8 compare on safety and coding benchmarks, Claude Sonnet 5 vs Opus 4.8 has the side-by-side breakdown.
Cyber classifiers on Opus 5 trigger around 85% less often than on Fable 5. Flagged requests in Claude.ai, Claude Code, and Claude Cowork fall back to Opus 4.8 by default. Enterprises already in Anthropic's Cyber Verification Program get a less-restricted build immediately.
Two API Beta Features Ship Alongside the Model
Anthropic shipped two beta updates with the launch. Mid-conversation tool changes let developers swap which tools Claude can call within a session without invalidating the prompt cache - a rebuild that previously forced a full context reset in many agentic setups. Automatic fallbacks on the API route safety-flagged requests to the best available model rather than returning a blocked error, which means production pipelines no longer fail silently when classifiers fire.
Fast Mode at 2.5x Speed, Biology Routing Gets an Upgrade
Fast mode runs Claude Opus 5 at roughly 2.5 times the default speed at twice the base price. In Claude Code, Fast mode draws from usage credits - the same billing system that currently applies to Fable 5 and Sonnet 5, detailed in Claude's full pricing breakdown. Access the model on the Claude API today with the claude-opus-5 identifier.
Biology teams get a direct upgrade alongside this launch. Biology-related requests that Fable 5 previously blocked now route to Opus 5 instead of Opus 4.8, since Opus 5 carries a similar safety profile with better capability. Opus 5 scores 10.2 percentage points higher than Opus 4.8 on organic chemistry tasks and 7.7 points higher on protein sequence work - numbers worth noting for teams running life sciences research with Claude Science.
Cybersecurity remains Fable 5's and Mythos 5's territory. At that performance-to-price ratio though, the harder question for most production teams is not whether to adopt Opus 5 - it's whether any coding or agent workload still needs Fable 5 at all. OpenAI, Google, and xAI have not published direct Frontier-Bench or CursorBench comparisons against Opus 5 yet.