Anthropic Opus 5 pricing undercuts Fable 5 as model tops index

Anthropic released Opus 5 on Friday, and the headline for enterprise buyers is Anthropic Opus 5 pricing: $5 per million input tokens and $25 per million output tokens. The company says the model gets close to Claude Fable 5’s frontier-level capability at half the token price, a direct appeal to customers watching inference costs rise.

Fable 5 is priced at $10 per million input tokens and $50 per million output tokens. OpenAI’s GPT-5.6 Sol is listed at $5 per million input tokens and $30 per million output tokens, putting Opus 5 level with Sol on input pricing and lower on output pricing. Anthropic did not disclose adoption targets, customer commitments or revenue expectations tied to the launch.

How much does Anthropic Opus 5 cost?

Opus 5 costs $5 per million input tokens and $25 per million output tokens, according to Anthropic. Token sticker prices do not tell the full cost story because different models can use different amounts of tokens to finish the same job.

Artificial Analysis puts the weighted average cost per Intelligence Index task at $2.75 for Fable, $2.03 for Opus 5 in max mode, $1.04 for GPT-5.6 Sol in max mode and $0.95 for Kimi K3. On that measure, Opus 5 is cheaper than Fable but still materially more expensive than the OpenAI and Kimi models cited in the benchmark.

Opus 5 also leads the Artificial Analysis Intelligence Index with a score of 61, one point above Fable. Anthropic says Opus 5 delivers much better performance than Opus 4.8 at the same cost as that predecessor. The company’s system card says the largest gains are in agentic coding, computer use and long-horizon knowledge work, and that Opus 5 is comparable to, or ahead of, Claude Fable 5 and Claude Mythos 5 in some cases.

What changed for developers using Claude Code?

Anthropic has reduced the size of the model’s system prompt, which should leave more usable context for customers. Claude Code engineer Thariq Shihipar said the company removed 80 percent of the Claude Code system prompt for the latest models.

That change affects how users should write prompts, skills and CLAUDE.md files. Shihipar said users may need to simplify those assets, and pointed to a command called claude doctor that can automatically optimize some prompts and skills.

There is a cost tradeoff in usage behavior. Anthropic warns that Opus 5’s default user-facing answers are longer than prior Opus models. Customers who do not want to pay for extra output tokens will need to tell the model directly to keep responses shorter.

What does Opus 5 change for security work?

Anthropic says Opus 5 should refuse fewer requests in cybersecurity and biology use cases. Its cyber classifiers are described as less restrictive than Fable 5’s: Anthropic says they allow source-code vulnerability discovery while blocking binary-based vulnerability scanning, penetration testing and exploit generation.

On the OSS-Fuzz benchmark for finding vulnerabilities in open source code, Opus 5 scored 80 percent, slightly above Mythos at 79.4 percent. It was much weaker than Mythos at turning those findings into working exploits, succeeding in 4 of 14 exploitation attempts compared with Mythos at 13 of 14.

Anthropic is also rolling out an automated API fallback option. Requests judged too risky for Opus 5 or Fable 5 can be sent to a less capable model instead of being rejected outright. Anthropic calls Opus 5 its most aligned model to date. For many businesses, the more concrete sales point may be that the model comes without a data retention requirement.

This story draws on original reporting from The Register.