Claude Opus 5: Benchmarks, Pricing, and Full Guide (July 2026)

Last updated: July 25, 2026

Claude Opus 5 is here. Anthropic released it on July 24, 2026 — a step-change improvement for the Opus tier that comes close to the frontier intelligence of Claude Fable 5 at half the price.

Quick answer: Claude Opus 5 costs $5/$25 per million input/output tokens (same as Opus 4.8), has a 1M-token context window, 128K max output, and a May 2026 knowledge cutoff — the most current of any Claude model. On Frontier-Bench v0.1 it more than doubles Opus 4.8’s score. On CursorBench 3.2 it performs within 0.5% of Fable 5’s peak at half the cost. On OSWorld 2.0 it surpasses Fable 5’s best result at a third of the cost. It is the new default on Claude Max and the strongest model on Claude Pro.

What is Claude Opus 5?

Claude Opus 5 is Anthropic’s newest and most capable Opus-class model. It is positioned as the everyday premium model — near-frontier model intelligence that you can use daily without Fable 5’s price or data-retention constraints.

Anthropic’s positioning is clear: Opus 5 is the model you reach for when you need strong intelligence on complex coding, agents, and professional work — and Fable 5 is for when you need the absolute ceiling regardless of cost.

Benchmarks: Opus 5 vs the field

Opus 5’s benchmark story is not incremental. On several evaluations it does not just beat Opus 4.8 — it changes the ranking.

Coding and software engineering

Three patterns stand out:

1. Frontier-Bench is a generational jump. Opus 5 more than doubles Opus 4.8’s score and surpasses all other models. This is the clearest evidence that Opus 5 is not an incremental update.
2. CursorBench nearly matches Fable 5. At max effort, Opus 5 is within 0.5% of Fable 5’s peak — at half the cost per task. At high, xhigh, and max effort, it achieves greater performance per dollar than any other model.
3. The cost-efficiency story is the real headline. On CursorBench, FrontierCode, and OSWorld 2.0, Opus 5 matches or approaches Fable 5’s results at a fraction of the cost.

Knowledge work, reasoning, and agents

Science and specialized domains

Opus 5 is a meaningful improvement over Opus 4.8 for scientific research. It outperforms Opus 4.8 on every life sciences evaluation, including:

• Organic chemistry: +10.2 percentage points over Opus 4.8 on inferring molecular structures from spectroscopy data
• Protein function prediction: +7.7 percentage points on predicting how sequence variations affect function
• Structural biology and bioinformatics: Consistent improvements across the board

Opus 5 is now Anthropic’s most capable generally available model for scientific research, since Fable 5’s biology safeguards are more restrictive.

Visual output quality

Opus 5 produces significantly stronger visual outputs — interactive artifacts, 3D work, animations, and data visualizations. Anthropic demonstrated a wind tunnel simulation and an interactive cell illustration, both built entirely by Opus 5. Customer feedback from Lovable and Gamma confirms the jump: “the best animations, games, and 3D work we have seen from an Opus model.”

Treat vendor-published benchmarks as directional. Validate on your own workload before committing budget.

What Opus 5 improved over Opus 4.8

This is not a minor version bump. The jump from Opus 4.8 to Opus 5 is the largest in the Opus line since Opus 4.5.

Customer feedback highlights a recurring theme: Opus 5 behaves more like a careful professional than a text generator. It verifies assumptions, cross-checks results by independent methods, and stays on track through long multi-step analyses.

Claude Opus 5 pricing

Opus 5 costs the same as Opus 4.8 — a significant value improvement given the capability jump.

Sonnet 5 introductory pricing runs through August 31, 2026.

The cost strategy: Opus 5 is the sweet spot for most production work. Use Fable 5 only when the task is hard enough that the remaining benchmark gap justifies 2x the token price. Use Sonnet 5 for high-volume work where near-Opus quality is sufficient.

New with Opus 5: Fast mode runs at approximately 2.5x the default speed at twice the base price. Available on the Claude Platform and through usage credits in Claude Code.

Claude Opus 5 vs Fable 5: when to use which

This is the most common decision for teams upgrading from Opus 4.8.

Rule of thumb: Start with Opus 5. Escalate to Fable 5 only when Opus 5 fails on your specific workload and the cost of failure exceeds the token premium.

For the detailed Fable 5 breakdown, see the Claude Fable 5 guide. For the previous-generation comparison, see Fable 5 vs Opus 4.8.

Claude Opus 5 vs Sonnet 5: the practical choice

For most teams, the real decision is Opus 5 vs Sonnet 5.

Sonnet 5 is the default recommendation for everyday work. Opus 5 is worth the premium when accuracy on complex, multi-step tasks justifies the cost — especially agentic coding, enterprise workflows, and scientific research.

For the full Sonnet 5 breakdown, see the Claude Sonnet 5 guide.

Claude Opus 5 vs the competition

Opus 5’s price-to-performance ratio is the story. At $5/$25 per MTok, it matches or approaches Fable 5 (and by extension GPT-5.6 Sol) on most benchmarks while costing half as much. GPT-5.6 Sol costs $5/$30 per MTok on short-context calls but applies a 2x input surcharge above 272K tokens — Opus 5’s 1M context has no such surcharge.

For the detailed head-to-head, see GPT-5.6 Sol vs Claude Fable 5.

Safety and safeguards

Opus 5 is Anthropic’s most aligned model to date. On the automated behavioral audit, it scores 2.3 on overall misaligned behavior — the lowest of recent Claude models. It adheres to Claude’s Constitution better than Opus 4.8, Sonnet 5, or Fable 5, exhibits the lowest rates of deceptive behavior, and is the least susceptible to being tricked into misuse.

Cybersecurity

Opus 5’s cyber classifiers (guardrails) are proportionally less restrictive than Fable 5’s — they intervene approximately 85% less often. The classifiers allow vulnerability discovery in source code but block binary-based vulnerability scanning, penetration testing, and exploit generation.

On OSS-Fuzz, Opus 5 is close to Mythos 5 at finding vulnerabilities but substantially behind on exploiting them. Flagged requests in Claude.ai, Claude Code, and Claude Cowork fall back to Opus 4.8 by default. API users can enable automatic fallbacks.

Members of the Cyber Verification Program (CVP) get access to a version with fewer restrictions.

Biology

Opus 5 has similar safeguards to Opus 4.8, making it the most capable generally available Claude model for scientific research. It still shows limitations on long-running autonomous research tasks, where Mythos 5 remains stronger. Biology-related requests blocked on Fable 5 now route to Opus 5 rather than Opus 4.8.

Data retention

Opus 5 has no data retention requirements for general access — consistent with prior Opus models. This is a significant enterprise advantage over Fable 5, which requires 30-day retention.

How to access Claude Opus 5

New API features launched alongside Opus 5

• Mid-conversation tool changes (beta): Change which tools Claude can use within a conversation without invalidating the prompt cache.
• Automatic fallbacks (beta): Requests flagged by safety classifiers on Opus 5 or Fable 5 automatically route to another model instead of being blocked.

The full Claude model lineup in July 2026

Legacy models still available

If you are still on Opus 4.8 or earlier, the upgrade to Opus 5 is free in price and massive in capability.

Which Claude model should you use?

The best metric is cost per successful task, not price per token. Opus 5’s efficiency gains — fewer retries, fewer tokens, less latency — mean it can be cheaper in practice than a lower-priced model that needs more iterations.

FAQ