Anthropic's Claude Opus 5 delivers near-Fable 5 performance at half the token price

2026-07-25 · Source: The Decoder · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, medium

Summary

Anthropic has released Claude Opus 5, its new flagship model, on July 25, 2026, delivering near-Fable 5 performance at half the token price. Positioned as a cheaper alternative, Opus 5 costs \$5 per million input tokens and \$25 per million output tokens, compared to Fable 5's \$10 and \$50 respectively. The model excels in agentic coding and knowledge work, leading benchmarks like Frontier-Bench v0.1 with 43.3% in agentic terminal coding and GDPval-AA v2 with an Elo score of 1,861. Notably, Opus 5 scores 30.2% on ARC-AGI-3 for novel problem-solving, nearly four times higher than GPT-5.6 Sol's 7.8%. It features a 1 million-token context window, can check and improve its own work, and build tools through code. Opus 5 also shows improved visual output generation and analysis, with cyber filters triggering 85% less often than Fable 5's.

Key takeaway

For AI Scientists or Machine Learning Engineers evaluating large language models for agentic coding or knowledge work, you should consider Claude Opus 5. Its competitive benchmark performance, especially on ARC-AGI-3 and agentic coding, combined with half the token price of Fable 5, presents a compelling cost-effective option. Experiment with its "low" or "medium" effort settings for general tasks to optimize cost, reserving "xhigh" for complex coding.

Key insights

Claude Opus 5 offers competitive performance against Fable 5 and GPT-5.6 Sol at significantly reduced token costs.

Principles

• Higher effort settings can reduce benchmark scores.
• Model self-correction improves complex task completion.
• Cost-performance trade-offs are crucial for model selection.

Method

Opus 5 can iterate to improve its own work and dynamically build tools via code, such as a computer vision pipeline for 3D model creation from drawings, demonstrating advanced agentic capabilities.

In practice

• Use "low" or "medium" effort settings for cost efficiency.
• Start with "xhigh" for coding and agentic tasks.
• Utilize automatic fallbacks for blocked API requests.

Topics

• Claude Opus 5
• Large Language Models
• Agentic AI
• Token Pricing
• AI Benchmarking
• Code Generation
• Knowledge Work

Best for: CTO, VP of Engineering/Data, AI Engineer, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• [AINews] Claude Opus 5: Fable-level performance at Opus price (half Fable) — Claude Opus 5 demonstrates advanced agentic coding and tool-use, challenging the adequacy of current aggregate benchmarks.
• Anthropic's Claude Opus 4.7 — Claude Opus 4.7 enhances agentic coding, computer vision, and knowledge work, but introduces tokenization and long-context performance tradeoffs.
• Anthropic's Opus 5 surprise — Claude Opus 5 delivers Fable 5-tier intelligence at half the cost, setting new benchmarks in agentic tasks and complex problem-solving.
• Anthropic's Claude Opus 5 costs well below Fable 5 while matching or beating it across most benchmarks — Claude Opus 5 leads benchmarks in capability and cost-efficiency, but its high hallucination rate demands careful application.
• Claude Opus 5 arrives with near Fable performance at half the price — Claude Opus 5 delivers near Fable 5 performance at half the cost, optimizing token use for complex agentic and knowledge work.
• not much happened today — Sonnet 5 enhances agentic capabilities and coding performance, establishing a new production workhorse for complex workflows.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by The Decoder.