Claude Opus 5: The Claude 5 Family Is Complete — Fable-5 Power at Half the Price

Claude Opus 5: The Claude 5 Family Is Complete — Fable-5 Power at Half the Price

On 24 July 2026, Anthropic released Claude Opus 5, and with it the Claude 5 family is complete: Fable 5 at the top, Sonnet 5 for the speed-and-intelligence balance, and now Opus 5 in the middle — the model Anthropic positions as its default choice for complex agentic coding and enterprise work. The headline: near–Fable 5 capability at half the price, and the very same price developers already paid for Opus 4.8.

Claude Opus 5 in brief

Opus 5 is the direct upgrade to Opus 4.8, built to be a strong everyday assistant: it works through a hard problem end to end without much back-and-forth, and handles both coding and general office work. The official numbers from Anthropic:

• Model ID: claude-opus-5
• Pricing: 5 dollars per million input tokens, 25 per million output — identical to Opus 4.8, and half of Fable 5.
• Context window: 1M tokens (≈ 555k words).
• Max output: 128k tokens.
• Reliable knowledge cutoff: May 2026 — newer than Opus 4.8 and Fable 5 (January 2026).
• Available today on Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry.

Why the Claude 5 family is now complete

Anthropic began its fifth generation on 9 June 2026 with Fable 5 (its most capable model, for long-running agents), followed by Sonnet 5 (the best speed-and-intelligence balance). The heart of the lineup — the Opus tier — stayed on the fourth generation until today. With Opus 5, all three primary tiers are now on the 5 generation:

Model	Position	Input / Output per 1M tokens	Context
Fable 5	Highest intelligence, long-running agents	$10 / $50	1M tokens
Opus 5	Agentic coding & enterprise work	$5 / $25	1M tokens
Sonnet 5	Best speed-and-intelligence balance	$3 / $15	1M tokens
Haiku 4.5	Fastest and cheapest	$1 / $5	200k tokens

(Haiku is still on 4.5, but the three top tiers — Fable, Opus, and Sonnet — are all on the 5 generation now.)

What changed from Opus 4.8

This is not a cosmetic bump. Anthropic's published comparisons of Opus 5 vs its predecessor:

• Coding: on Frontier-Bench v0.1, Opus 5 more than doubles Opus 4.8's performance — at a lower cost per task.
• Science: 10.2 percentage points higher on organic chemistry, and 7.7 points higher on protein-prediction tasks.
• Fresher knowledge: May 2026 cutoff vs January 2026 for 4.8.
• Same price: all of this without raising the cost above Opus 4.8.

What changed from Fable 5 (the flagship)

Fable 5 remains the reference for the hardest possible tasks, but Opus 5 gets strikingly close at half the cost — and beats it in several practical areas:

• Nearly matches its coding: on CursorBench 3.2, within 0.5% of Fable 5 — at half the cost per task.
• Wins on computer use: on OSWorld 2.0, it surpasses Fable 5's result at roughly one-third the cost.
• Better at business workflows and agentic research.
• Far fewer restrictions: it refuses and interrupts about 85% less often than Fable 5, making it better for broad, everyday deployment.
• Where Fable 5 still leads: the most advanced tasks such as cybersecurity exploit development and advanced biology (a deliberate safety gap).

Notably, Anthropic says that even at minimal effort, Opus 5 succeeds at more business-workflow tasks than other models running at high effort.

Opus 5 vs competitors, by the numbers: GPT-5.5, Gemini 3.1, and Grok 4

The real competition today is between OpenAI's GPT family (latest is GPT-5.5, launched April 2026), Google's Gemini (version 3.1), xAI's Grok 4, and DeepSeek. Anthropic's official Opus 5 figures were framed against the "next-best model" without naming it, so we complete the picture with independent numbers. Important caveat: because Opus 5 launched today, direct independent head-to-heads for it aren't in yet. The table below benchmarks Opus 4.8 (Opus 5's direct predecessor) against the field — and Opus 5 builds on and exceeds that baseline.

Benchmark	Claude Opus 4.8	GPT-5.5	Gemini 3.1 Pro
SWE-bench Verified (real-world coding)	88.6%	≈ 80%	≈ 82%
SWE-bench Pro (harder coding)	69.2%	≈ 64%	Lower
Terminal-Bench 2.1 (terminal agent)	74.6%	78.2%	—
OSWorld (computer use)	83.4%	78.7%	—
GDPval-AA (knowledge work)	1890	1769	1314
Context window	1M tokens	256k	1M tokens
Input price / 1M tokens	$5	Varies	$2 (under 200K)

The read: the Opus line leads on real-world coding, computer use, and knowledge work, while GPT-5.5 edges ahead on some terminal tasks. On hard math, Anthropic's flagship Fable 5 tops FrontierMath at 87.0% vs 85.3% for GPT-5.5.

And Opus 5 itself? Anthropic says it beats the "next-best model" by roughly 3× on the abstract-reasoning benchmark ARC-AGI 3, and by about 1.5× on Zapier AutomationBench (agentic automation) for the same cost per task. On an automated behavioral audit it scored 2.3 on misaligned behavior — the lowest (best) among the compared models. In other words, Opus 5 raises the already-leading Opus 4.8 baseline above, not the reverse.

Key new features

• The "effort" switch: choose maximum intelligence, or conserve tokens for faster, cheaper results — per task.
• Fast mode: up to 2.5× the default speed (at double the price) for latency-sensitive work.
• Mid-conversation tool changes (beta) for more flexible agent workflows.
• Automatic model fallbacks (beta): switch to another model automatically when needed.
• 1M-token context for large codebases and documents in one pass.

What this means for businesses

The core of the news for any company: capabilities once reserved for the priciest model are now available at half the cost. That lowers the price of building AI assistants, automating operations, and internal coding tools — without sacrificing quality. And because Opus 5 refuses and interrupts less, with fresher knowledge, it's a more practical choice for broad deployment to users and staff.

The takeaway

With Opus 5 completing the Claude 5 family, "front-of-pack performance at a sensible price" is now available to everyone: results that rival GPT-5.5 and Gemini 3.1 and approach Fable 5, at half the top-tier cost, with fresher knowledge. For business owners, it's a good moment to re-evaluate what can be automated and built with AI.

Sources

• Anthropic — Claude Opus 5 announcement: https://www.anthropic.com/news/claude-opus-5
• Anthropic — Models overview (official specs & pricing): https://platform.claude.com/docs/en/about-claude/models/overview
• Independent comparison figures (Opus 4.8 vs GPT-5.5 and Gemini 3.1 Pro): hundredtabs.com, bitsminds.com, lmcouncil.ai