Anthropic released Opus 5 on Friday with a very pointed claim: it comes close to the frontier intelligence of Claude Fable 5, but at half the price. In concrete terms, Anthropic (per The Register) priced Opus 5 tokens at $5 per MTok input and $25 per MTok output, compared with Fable 5 at $10 per MTok input and $50 per MTok output. If you are an enterprise trying to keep AI usage from turning into a bottomless line item, that is not a marketing tweak. It is a budget re-rate.

But the “half the price” headline gets sharper when you look at what it costs to actually complete work, not just to buy tokens. The piece cites Artificial Analysis weighted average cost per Intelligence Index task: $2.75 for Fable, $2.03 for Opus 5 (max), $1.04 for GPT-5.6 Sol (max), and $0.95 for Kimi K3. In other words, Opus 5 is not just cheaper in rate terms. It is also materially better on task economics versus Fable. And in the same source, Artificial Analysis places Opus 5 at 61 in its Intelligence Index, one point ahead of Fable.

So why does this matter beyond a spreadsheet? Because enterprise AI pricing is often treated like a simple multiplication problem, when real usage is a messy mix of prompt design, response length, and how many attempts a system needs to get a usable output. Anthropic itself hints at that complexity with an important tradeoff: Opus 5 may produce wordier responses by default. The Register reports Anthropic cautions that Claude Opus 5's default user-facing responses run longer than prior Opus models. The practical implication for operators is simple. If you want the cost advantage from Opus 5 pricing to show up in reality, you may need to prompt for response length explicitly.

There is also a product reason for that shift. Opus 5 arrives with a “system prompt on a diet.” Claude Code engineer Thariq Shihipar is quoted saying Anthropic removed 80 percent of the Claude Code system prompt for its latest models. That is not just trivia. It changes how guidance is delivered, which affects how prompts, skills, and CLAUDE.md files should be structured. Shihipar is quoted explaining that across your system prompt, skills, and CLAUDE.md files, you may need to simplify “just like we did,” and that a command called claude doctor can help automatically optimize some prompts and skills. If you have a team that already built internal prompting frameworks for Opus 4.8-era models, this is the kind of update that can quietly break your “known good” prompt templates, or force a re-tuning of your workflows.

On performance, the source paints Opus 5 as a step up for work that needs more context and longer arcs. It notes Anthropic’s claim that Opus 5 provides greatly improved performance for the same cost as its predecessor, Opus 4.8. It also cites that Opus 5 scores close to Mythos on finding vulnerabilities in open source code, with an 80 percent OSS-Fuzz benchmark score versus 79.4 percent for Mythos. The same benchmark picture flips for weaponization: Opus 5 succeeded in 4 out of 14 exploitation attempts compared to 13 out of 14 for Mythos. Put plainly, Opus 5 is strong at identifying issues, but weaker at the next step of turning those issues into exploits. That aligns with the operational concern many enterprises have had for a while: capability is one axis, control is another.

Control is exactly where Anthropic is leaning in with two additional changes. First, Opus 5’s system card says it is “substantially stronger than Claude Opus 4.8 across the board,” with the largest gains in agentic coding, computer use, and long-horizon knowledge work, and it is described as comparable to, or in some cases ahead of, Claude Fable 5 and Claude Mythos 5. Second, the source reports a safety-related pricing unlock of sorts: Anthropic says Opus 5’s cyber classifiers are proportionally less restrictive than those on Fable 5. The effect is that Opus 5 can find vulnerabilities in source code, while blocking “binary-based” vulnerability scanning, penetration testing, and exploit generation. If you are deploying an AI for security engineering, that could mean fewer dead ends during routine code review and scanning workflows.

Finally, there is a procurement and governance lever that CFOs will actually feel. The Register says Anthropic is making life easier for users frustrated by refusals by rolling out an automated fallback function on its API. Requests deemed too dangerous to entrust to Opus 5 or Fable 5 can be routed to a less capable model instead of being blocked outright. And perhaps the most enterprise-friendly point in the entire report: Opus 5 comes with no data retention requirement, described as “a more compelling selling point for businesses.” In an environment where data handling clauses and model usage policies can take weeks to negotiate, the difference between “no retention requirement” and “we need legal to approve every exception” can be the difference between a pilot and a permanent program.

Zoom out and you can see why Opus 5 is a board-level story, not just a model release. Token costs are already a strategic constraint for adoption, and refusal behavior is already a workflow constraint for teams building agents. Anthropic’s combination of cheaper unit economics versus Fable 5, higher Intelligence Index positioning, an announced approach to prompt simplification, and a governance mechanism that falls back rather than blocks could shift what procurement expects from “frontier-ish” models. If you are evaluating competitors or negotiating model contracts for cybersecurity, coding, or long-horizon business automation, Opus 5 is a reminder that the winning product is often the one that controls cost, controls risk, and still gives users outputs they can ship.