Anthropic released Claude Fable 5 on June 9, 2026, as its most capable model yet for coding and complex reasoning tasks, but it costs exactly twice as much per token as the previous flagship Claude Opus 4.8. The new model scores 80.3% on SWE-Bench Pro, a rigorous test of real-world software engineering ability, compared to Opus 4.8's 69.2%. For teams building AI agents and automated coding systems, this 11-point jump represents a meaningful leap forward. However, the pricing restructure forces developers to actively choose between tiers rather than automatically inheriting improvements.

What Makes Claude Fable 5 Different From Opus 4.8?

Anthropic introduced an entirely new tier called "Mythos" to house Fable 5, marking the first time the company has positioned a model above Opus rather than simply updating it. This shift reflects a fundamental change in how Anthropic is organizing its product line. Fable 5 was built specifically for long-horizon agentic work, meaning it excels at tasks where an AI system must plan multiple steps ahead, use tools in sequence, and manage its own working memory across hours-long coding or research sessions.

The two models share identical context windows of 1 million tokens, roughly equivalent to processing 100,000 words at once. Both can output up to 128,000 tokens per response. The real differences emerge in performance and cost. Fable 5 costs $10 per million input tokens and $50 per million output tokens, while Opus 4.8 costs $5 and $25 respectively.

How Do the Benchmark Scores Compare Across Different Tasks?

The performance gap depends heavily on what task you're measuring. Fable 5 dominates on coding benchmarks, but the picture becomes more nuanced when examining other types of reasoning. Here's how the two models stack up across key metrics:

• SWE-Bench Pro (coding tasks): Fable 5 achieves 80.3% versus Opus 4.8's 69.2%, an 11.1-point advantage on professionally verified software engineering problems designed to resist memorization.
• Tau-Squared Bench (multi-step tool use): Fable 5 scores 98.5%, demonstrating exceptional reliability at completing complex, multi-step tasks that require calling different tools in sequence.
• GPQA Diamond (PhD-level science reasoning): Opus 4.8 actually edges out Fable 5 with 93.6% versus 92.6%, suggesting Fable 5 was tuned for coding throughput rather than pure scientific reasoning.

This pattern reveals Anthropic's engineering choices. Fable 5 was optimized for the specific demands of agentic systems and code generation, not for general-purpose reasoning across all domains. Teams working on scientific research or theoretical problem-solving might find Opus 4.8 remains the better choice, despite being the older model.

What's New About Fable 5's Safety Features?

Fable 5 introduces a novel safety routing mechanism that automatically hands off sensitive requests to Opus 4.8 for handling. This feature exists because Anthropic also released a sibling model called Claude Mythos 5, which shares Fable 5's raw capabilities but removes the safety classifiers. Mythos 5 remains restricted to government and cybersecurity customers, while Fable 5 is available to all Pro, Team, Enterprise, and API users. The safety routing explains why Mythos 5 stays locked down; without classifiers, it cannot automatically identify and redirect sensitive prompts.

Steps to Evaluate Whether Fable 5 Fits Your Workflow

For teams considering whether to migrate from Opus 4.8 to Fable 5, several practical factors should guide the decision:

• Coding intensity: If your production workload involves heavy software engineering tasks, automated code generation, or debugging, the 11-point improvement on SWE-Bench Pro may justify the doubled cost. Teams running lighter coding loads might see diminishing returns.
• Agentic complexity: If you're building systems where the AI must manage multiple tool calls, maintain state across long sessions, and adapt its approach mid-task, Fable 5's 98.5% score on multi-step tool use suggests significantly higher reliability than Opus 4.8.
• Science and reasoning workloads: If your primary use case involves PhD-level reasoning, theoretical analysis, or domain-specific problem-solving, Opus 4.8's slight edge on GPQA Diamond indicates it may remain competitive despite being the older model.
• Budget constraints: The exact doubling of per-token costs means a migration to Fable 5 will double your API spend for equivalent usage. Teams on tight budgets should calculate the expected improvement in output quality and task success rates before committing.
• Context window needs: Neither model expanded the context window, so if you need to process documents larger than 1 million tokens, neither Fable 5 nor Opus 4.8 will solve that problem. Competitors like DeepSeek V4 may offer better options for document-heavy workflows.

How Does Fable 5 Fit Into Anthropic's Broader Lineup?

The June 2026 release cycle was unusually compressed for Anthropic. Opus 4.8 launched on May 28 as the flagship. Eleven days later, Fable 5 and Mythos 5 arrived and demoted Opus to the middle tier. Then on June 30, Anthropic released Claude Sonnet 5 as the new default free and Pro-tier model, completing a three-model refresh in just over a month.

This restructuring reflects growing competition in the AI market. Anthropic now offers Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens, making it the cheapest entry point. Opus 4.8 sits in the middle at $5 and $25. Fable 5 occupies the premium tier at $10 and $50. The company is no longer simply iterating on a single flagship model; it's building a tiered ecosystem where different teams choose different models based on their specific needs and budgets.

Search interest in comparisons like "Claude vs Gemini" and "Claude vs ChatGPT" spiked through June as developers tried to understand where Fable 5 fit relative to competitors. The benchmark data suggests Fable 5 is Anthropic's strongest coding and agentic model to date, but it does not represent a breakthrough in context window size or general reasoning ability. For many teams, the decision to upgrade will hinge on whether their specific workload matches Fable 5's tuning priorities.