Anthropic Launches Claude Fable 5 to the Public with 95% on SWE-bench Verified and Charges Double that of Opus 4.8
The first model in the Mythos class available to the public arrives at the API for $10 per million input tokens and $50 for output, achieving 80.3% on SWE-bench Pro compared to 58.6% for GPT-5.5.
On Tuesday (9), Anthropic released Claude Fable 5, the first version of the Mythos class made available to the general public, significantly altering the frontier pricing table. The model charges $10 per million input tokens and $50 per million output tokens via API, double that of Claude Opus 4.8 ($5 / $25) and double that of GPT-5.5's input price ($5 / $30). Dario Amodei stated in the announcement that Fable 5's capabilities "exceed those of any model we've previously made publicly available."
The benchmark justifying the premium is coding. On the SWE-bench Verified, Fable 5 scores 95%; on the SWE-bench Pro, a stricter and independent metric, it achieves 80.3% compared to 69.2% for Opus 4.8 and 58.6% for GPT-5.5. The context window extends to 1 million tokens, with output of up to 128 thousand, a range in which Opus 4.8 already led with a score of 68.1% on GraphWalks BFS against 45.4% for OpenAI. In the Terminal-Bench, the equation reverses: GPT-5.5 wins with 82.7% compared to 80.5% for Fable 5, reminding us that the race for frontier models is no longer a straightforward queue.
Distribution: Bedrock, Vertex, and Copilot on Day Zero
Anthropic learned the lesson from Opus 4 and secured simultaneous distribution across three hyperscalers and the largest IDE in the market. Fable 5 is now available on Amazon Bedrock and the Claude Platform on AWS, on Google’s Vertex AI as a partner model in Model Garden, and within GitHub Copilot for Pro+, Max, Business, and Enterprise plans, with support for VS Code, Visual Studio, JetBrains, Xcode, Eclipse, CLI, github.com, and Copilot Mobile. Copilot Enterprise administrators need to manually activate the Fable 5 policy, which is off by default. Until June 22, the model is included at no additional cost in Pro, Max, Team, and Enterprise plans per seat, before transitioning to a usage credit system.
The security architecture comes with caveats. Anthropic retains prompts and responses for up to 30 days to feed classifiers that detect abusive usage in areas such as offensive cybersecurity, biology, and chemistry. When a query falls into one of these "high-risk" domains, Fable 5 automatically falls back to Opus 4.8, a less capable model, and responds from there. For regulated customers in the EU and UK, mandatory retention opens a difficult conversation with privacy areas: the zero data retention standard that Anthropic markets for the rest of the line does not apply here.
For Whom the Doubled Price Makes Sense
The verdict is not uniform. In short tasks, simple prompts, or low-complexity queries, doubling the ticket for a marginal improvement burns budget without ROI. The comparative analysis published by Finout and ofox.ai converges: the premium for Fable 5 pays off when the context is long, the task requires multiple steps, and the work builds along an agentic chain. For everything else, Opus 4.8 and Sonnet 4.6 deliver better cost efficiency.
Market reading unfolds along two fronts. In the United States, where OpenAI still dominates the developer-facing API traffic and GPT-5.5 Pro costs $30 / $180, Anthropic positions itself as the vendor of choice for reasoning-intensive software engineering workloads, a segment that Aiman Ezzat's team at Capgemini and Julie Sweet's team at Accenture have reported as the highest demand in consulting. In India, where TCS, Infosys, and Wipro have just crossed the mark of 300,000 activated Microsoft 365 Copilot seats, the arrival of Fable 5 on Bedrock and Copilot accelerates the pressure to rewrite the offshore delivery economy: if a team of 12 engineers delivers in GPT-5.5 what ten do in Fable 5 faster and with less rework, the utilization arithmetic collapses.
The most relevant signal from the announcement is not in the pricing table, but in the positioning. By reserving the "Mythos" label for the full-capacity version that remains restricted to a list of about 50 developers, Anthropic transforms controlled capability into a product. The next frontier fight will not be about who has the strongest model, but rather about who decides how much of that model the market can purchase.