Claude Opus 5: everything you need to know about Anthropic’s new near-frontier model

On July 24, 2026, Anthropic launched Claude Opus 5, a near-frontier model designed for daily production use . The pitch is simple: a model that gets close to its most powerful publicly available system, Claude Fable 5, at half the price . Here is a complete, sourced breakdown of the launch—pricing, benchmark performance against Fable 5 and Opus 4.8, safety findings from its system card, and a major upgrade to Claude's voice mode.

Pricing and positioning in the 2026 lineup

Opus 5 sits in a carefully constructed pricing ladder. It replaces Opus 4.8 (launched May 28, 2026) as the new high-end accessible workhorse, sitting below Claude Fable 5 (Anthropic's most capable public model, launched June 9, 2026) and above Claude Sonnet 5 (launched June 30, 2026, now the default free/pro model) .

Model	Input (per million tokens)	Output (per million tokens)
Sonnet 5	$2 (intro) → $3	$10 (intro) → $15
Opus 5	$5	$25
Opus 4.8	~$3–$4	~$15
Fable 5	$10	$50

Opus 5 is priced identically to its predecessor, Opus 4.8 . Anthropic says the new model provides "greatly improved performance for the same cost" as Opus 4.8 .

A key differentiator: Opus 5 introduces a low/medium/high "effort toggle" that lets users trade cost for capability per request — a first for the lineup . In max-effort mode, Opus 5 scores 61 on the Artificial Analysis Intelligence Index .

Benchmark performance: Opus 5 vs. Fable 5 vs. Opus 4.8

Opus 5 vs. Fable 5: Anthropic says Opus 5 outperforms Fable 5 on coding and knowledge work evaluations while costing half as much . CNBC reports Opus 5 "rivals" Fable 5 performance at roughly half the price . MacRumors characterizes it as offering "near Fable 5 performance" . The Times of India notes that according to Anthropic's own data, Opus 5 is better than Fable 5 on 10 out of 14 metrics .

It is important to clarify the nuance: on the hardest agentic-coding benchmarks (e.g., SWE-Bench Pro), the Fable 5 scores 80.3% vs. Opus 4.8 at 69.2% . Opus 5 closes that gap significantly . On the Every team's Senior Engineer benchmark — designed to be hard for human senior engineers — Fable 5 scores 91/100 and Opus 4.8 scored 63/100; Opus 5 lands near Fable 5's level .

Opus 5 vs. Opus 4.8: Opus 5 scores higher across agentic coding, novel problem solving, knowledge work, computer use, and multidisciplinary reasoning — described as a "meaningful improvement" . On Frontier-Bench v0.1, Anthropic says Opus 5 "surpasses all other models, and more than doubles Opus 4.8's performance at a lower cost per task" .

The headline takeaway: Opus 5 delivers roughly Fable 5-level intelligence for most tasks at half the price, making it Anthropic's most cost-effective high-end offering .

System card and safety findings

The Claude Opus 5 System Card reports the following key findings :

• Usage policy and user wellbeing: Performs comparably to Opus 4.8 across evaluations for usage policy compliance, user wellbeing, child safety, and bias and integrity.
• Harmlessness: Maintained high harmless response rates on single-turn harmful requests.
• Over-refusal: Achieves among the lowest over-refusal rates on benign requests of any recent Anthropic model.
• Misalignment risk: The evaluation found no evidence of dangerous coherent misaligned goals, consistent with Opus 4.8-level alignment. The model is described as the "most aligned model to date" on Anthropic's automated behavioral audit, surpassing Sonnet 5, Opus 4.8, and even Mythos 5 on a variety of alignment evaluations .
• Cyber capabilities: Improved over Opus 4.7 but still well behind the restricted Mythos class .
• Honesty: Improved quite a bit across the board, especially agentic honesty .

Notably, Opus 5 shares the same underlying model family as the restricted Claude Mythos 5 but adds stronger safeguards for high-risk domains (cybersecurity, biology, chemistry, model distillation) .

Claude voice mode gains Opus and Sonnet

On July 23, 2026, Anthropic upgraded Claude's voice mode so paid users can now run spoken conversations on Opus or Sonnet models — previously limited to the lighter Haiku 4.5 .

Key details :

• Users can switch models mid-conversation between Haiku, Sonnet, and Opus.
• Voice mode now integrates with connected apps: Gmail, Google Calendar, Google Docs, Slack, Canva, web search.
• 10 non-English languages supported in beta.
• The interface defaults to whichever model the user last used in text chat, running its fastest variant to keep the exchange flowing.
• Free users remain on Haiku with one connected app.

The expansion unlocks a class of voice interactions that were impractical on Haiku, including tasks requiring extended reasoning chains and multi-step tool use handled entirely through speech .

What the research could not confirm

This briefing investigated three additional questions, but the available sources did not contain enough evidence to report specifics:

End of free Fable 5 access. Fable 5 was initially offered as a limited free-tier trial beginning June 9, 2026 . The exact date and terms of the free access ending could not be confirmed from the retrieved sources.

Expiration of the Claude Code usage limit boost. A temporary increase in Claude Code requests was likely offered around the Sonnet 5 or Fable 5 launch, but the specific expiration date was not documented in the captured sources.

Deloitte AI security platform built on Claude. No announcement about Deloitte launching an AI security platform on Claude was found in the search results.

The bottom line

Claude Opus 5 is a significant release that reshapes Anthropic's lineup. For most users and organizations, it offers the best balance of capability and cost: Fable 5–level intelligence for everyday production work at half the token price. Combined with the effort toggle for fine-grained cost control and a voice mode that now runs on the same powerful models, Opus 5 makes frontier-grade AI more accessible than ever.