AI and ML
New model matches flagship capabilities while eliminating data retention requirements
Anthropic unveiled its Opus 5 AI model on Friday, positioning it as a cost-efficient alternative that “comes close to the frontier intelligence of Claude Fable 5 at half the price.” The release addresses growing enterprise concerns over escalating token expenses.
Fable 5 is priced at $10 per million input tokens and $50 per million output tokens, whereas Opus 5 reduces those rates to $5 and $25 respectively. For comparison, OpenAI’s GPT-5.6 Sol sits at $5 per million input tokens and $30 per million output tokens. However, per-token pricing alone does not capture total cost of ownership, as the number of tokens required to complete a given task varies across models.
According to Artificial Analysis, the weighted average cost per Intelligence Index task stands at $2.75 for Fable, $2.03 for Opus 5 (max), $1.04 for GPT-5.6 Sol (max), and $0.95 for Kimi K3. On the Intelligence Index itself, Opus 5 leads with a score of 61, edging out Fable by one point.
“Claude Opus 5 provides greatly improved performance for the same cost as its predecessor, Opus 4.8,” Anthropic stated.
Opus 5 also offers expanded usable context. Anthropic has significantly reduced the system prompt overhead—Claude Code engineer Thariq Shihipar noted that 80 percent of the Claude Code system prompt was removed for the latest models. Consequently, the company’s guidance for constructing prompts, skills, and CLAUDE.md files has been updated. “Across your system prompt, skills, and CLAUDE.md files, you may need to simplify just like we did,” Shihipar advised, adding that a claude doctor command can assist in automatically optimizing prompts and skills.
One trade-off: Opus 5 tends toward verbosity. Anthropic cautions that “Claude Opus 5’s default user-facing responses run longer than prior Opus models’,” which can increase output token consumption. Customers seeking tighter responses are encouraged to specify length constraints explicitly in their prompts.
In security evaluations, Opus 5 matches the Mythos model on vulnerability discovery, scoring 80 percent on the OSS-Fuzz benchmark versus Mythos’s 79.4 percent. However, it lags significantly in exploit weaponization, succeeding in only 4 of 14 exploitation attempts compared to Mythos’s 13 of 14.
The model’s system card characterizes Opus 5 as “substantially stronger than Claude Opus 4.8 across the board, with the largest gains in agentic coding, computer use, and long-horizon knowledge work,” placing it on par with or ahead of both Claude Fable 5 and Claude Mythos 5 in several dimensions.
The model is expected to prove particularly valuable for cybersecurity and biology workloads due to reduced refusal rates. Anthropic explained that “Opus 5’s cyber classifiers are proportionally less restrictive than those on Fable 5. They allow Opus 5 to find vulnerabilities in source code, but block ‘binary-based’ vulnerability scanning (a method more likely to be associated with malicious actors), penetration testing, and exploit generation.”
To further mitigate disruption from safety refusals, Anthropic is introducing an automated fallback mechanism on its API. Requests deemed too sensitive for Opus 5 or Fable 5 will be routed to a less capable model rather than blocked outright.
Opus 5 is described as Anthropic’s “most aligned model to date,” reflecting the lowest propensity for guardrail violations. For enterprises, a more pragmatic advantage may be the absence of a data retention requirement.
