{"slug": "claude-opus-5-frontier-performance-at-half-the-price-of-fable-5", "title": "Claude Opus 5: Frontier Performance at Half the Price of Fable 5", "summary": "Anthropic launched Claude Opus 5 on July 24, 2026, offering near-Fable-5 intelligence at $5 per million input tokens, half the cost of Fable 5's $10, and beating Fable 5 on most benchmarks including SWE-bench Verified (96.0%) and Frontier-Bench (43.3%). The model introduces an effort toggle with five levels to control internal reasoning, and Anthropic recommends Opus 5 as the default for most production use cases, with Fable 5 still preferred for multi-day autonomous agents.", "body_md": "Anthropic launched **Claude Opus 5** on July 24, 2026, and the headline writes itself: you get near-Fable-5 intelligence at exactly half the price. That’s $5 per million input tokens versus Fable 5’s $10, with benchmark results that beat Fable 5 on most evaluations. If your team is currently running production agents on Fable 5, this is a release that warrants an immediate cost review.\n\n## What Changed From Opus 4.8\n\nAnthropic describes Opus 5 as a “full generation of gains” over its predecessor. The numbers bear that out. SWE-bench Multimodal jumps from 38.4% to 59.4%. Frontier-Bench — the benchmark for agentic terminal coding — comes in at 43.3%, ahead of Fable 5. On SWE-bench Verified, Opus 5 hits 96.0%. These aren’t incremental improvements; they’re the kind of numbers that warrant revisiting whatever workloads you ruled out for AI assistance a year ago.\n\nThe model also ships with a new **effort toggle** — an API parameter with five levels (low, medium, high, xhigh, max) that controls how much internal reasoning the model performs before responding. Higher effort means more thinking tokens, which bill as output tokens. The practical upside: low and medium effort on Opus 5 now outperform high effort on Opus 4.8, which means you can tune down effort on simpler tasks and cut your token bill without sacrificing quality.\n\n## Benchmarks That Matter\n\nHere’s how Opus 5 stacks up against Fable 5 on the metrics developers actually care about:\n\n| Benchmark | Claude Opus 5 | Claude Fable 5 |\n|---|---|---|\n| SWE-bench Verified | 96.0% | ~95% |\n| SWE-bench Pro | 79.2% | ~79.5% |\n| Frontier-Bench (agentic coding) | 43.3% | Lower |\n| GDPval-AA v2 | 1,861 | 1,747 |\n| BrowseComp (agentic search) | 90.8% | Lower |\n| SWE-bench Multimodal | 59.4% | — |\n\nThe one area where Fable 5 edges ahead — SWE-bench Pro — is by less than a single percentage point. For the cost of two Fable 5 calls, you can make two Opus 5 calls and run your evals twice. The math is not subtle.\n\n## When to Use Which Model\n\nAnthropic is direct about the remaining gap: Fable 5 retains an advantage for **multi-day autonomous agents** — workloads that require days of planning, self-correction across very long trajectories, and sustained judgment. If you’re building products where agents run unsupervised for 24+ hours, Fable 5 remains the recommendation.\n\nFor offensive cybersecurity and advanced biology research, neither Opus 5 nor Fable 5 is the top choice — that’s [Mythos 5](https://www.anthropic.com/news/claude-fable-5-mythos-5)’s territory.\n\nFor everything else — and that covers the vast majority of production use cases — Opus 5 is now the default recommendation. Anthropic put it on Claude Max as the default model at launch. The signal is clear.\n\n## Migrating and Getting Started\n\nThe model is available now under the ID `claude-opus-5`\n\nacross [Anthropic’s API](https://www.anthropic.com/news/claude-opus-5), Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. The context window is 1 million tokens with a 128K max output.\n\nIf you’re upgrading from Opus 4.8: same price, substantially better performance. Swap the model ID and run your evals. If you’re migrating from Fable 5: 50% cost reduction with comparable or better results on most workloads. The migration path is a one-line change.\n\nOne thing worth testing before you commit: the effort toggle. The [official Anthropic docs](https://platform.claude.com/docs/en/build-with-claude/effort) recommend starting at high and tuning down based on your evals. Tasks that required high effort on Opus 4.8 may now run well on medium — at meaningfully lower cost per call.\n\n## Bottom Line\n\nClaude Opus 5 is the most consequential Anthropic release for production developers this year. Not because it’s the most powerful model — that’s Mythos 5 — but because it puts frontier-class performance within reach at a price point that makes AI-powered workflows economically viable at scale. For most teams: switch to Opus 5, run your evals, and pocket the savings. Fable 5 is the call only if your workloads involve multi-day autonomous operation. Everything else belongs on Opus 5 now.\n\n*For full benchmark details and the system card, see CNBC’s coverage and the VentureBeat enterprise breakdown.*", "url": "https://wpnews.pro/news/claude-opus-5-frontier-performance-at-half-the-price-of-fable-5", "canonical_source": "https://byteiota.com/claude-opus-5-frontier-performance-at-half-the-price-of-fable-5/", "published_at": "2026-07-28 02:07:04+00:00", "updated_at": "2026-07-28 02:23:08.616359+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-products", "ai-tools", "ai-infrastructure"], "entities": ["Anthropic", "Claude Opus 5", "Fable 5", "Mythos 5", "Amazon Bedrock", "Google Cloud Vertex AI", "Microsoft Foundry", "Claude Max"], "alternates": {"html": "https://wpnews.pro/news/claude-opus-5-frontier-performance-at-half-the-price-of-fable-5", "markdown": "https://wpnews.pro/news/claude-opus-5-frontier-performance-at-half-the-price-of-fable-5.md", "text": "https://wpnews.pro/news/claude-opus-5-frontier-performance-at-half-the-price-of-fable-5.txt", "jsonld": "https://wpnews.pro/news/claude-opus-5-frontier-performance-at-half-the-price-of-fable-5.jsonld"}}