Anthropic released its Opus 5 AI model on Friday, stating that it approaches the frontier intelligence of its Fable 5 model while charging half the token price. Fable 5 prices tokens at $10 per million input and $50 per million output; Opus 5 prices tokens at $5 per million input and $25 per million output. OpenAI's GPT-5.6 Sol model lists prices of $5 per million input and $30 per million output. Artificial Analysis calculated weighted average costs per Intelligence Index task as $2.75 for Fable, $2.03 for Opus 5 (maximum), $1.04 for GPT-5.6 Sol (maximum) and $0.95 for Kimi K3. Opus 5 achieved an Intelligence Index score of 61, one point higher than Fable. Anthropic said Opus 5 improves performance over Opus 4.8 at the same cost and reduced its system prompt by removing 80 percent of the original content. The company noted that the trimmed prompt may require users to simplify system prompts, skills and CLAUDE.md files, and that a tool named claude doctor can help optimize prompts. Opus 5 responses may be longer than previous Opus models, and Anthropic advised users to request shorter outputs explicitly. The model scored 80 percent on the OSS-Fuzz benchmark for vulnerability discovery and completed four of fourteen exploitation attempts, compared with thirteen of fourteen for Mythos. Its system card states it is substantially stronger than Opus 4.8 across all metrics, with the largest improvements in agentic coding, computer use and long‑horizon knowledge work, and is comparable to or ahead of Fable 5 and Mythos 5. Anthropic said Opus 5’s cyber classifiers are less restrictive than those on Fable 5, allowing code vulnerability detection while blocking binary‑based scanning, penetration testing and exploit generation. An automated fallback function on the API now routes unsafe requests to a less capable model instead of blocking them outright. Anthropic described Opus 5 as its most aligned model to date and noted that it imposes no data retention requirement.