Anthropic Launches Claude Fable 5.1: Revolutionary AI Model Doubles Science Benchmarks, Slashes Costs by 75%

Anthropic Unveils Most Advanced AI Models for Coding and Knowledge Work

Anthropic has officially launched Claude Fable 5.1 and Claude Mythos 5.1, marking a significant leap forward in artificial intelligence capabilities for complex coding, scientific research, and long-horizon problem-solving tasks. Released on September 1, 2026, these frontier models represent Anthropic's most powerful systems to date, with Fable 5.1 achieving a groundbreaking 52.6% score on Terminal-Bench-Science 0.1—more than double the performance of its predecessor, Fable 5, which scored 24.7%.anthropic+4

The launch positions Anthropic in direct competition with other AI leaders, as Fable 5.1 also outperforms Claude Opus 5 (29.0%) and GPT-5.6 Sol (22.4%) on the same benchmark. Terminal-Bench-Science 0.1 evaluates agentic scientific research capabilities across 70 scientist-contributed workflows spanning life sciences, physical sciences, Earth sciences, mathematics, and engineering. Agents work in self-contained terminal environments and submit artifacts that are graded against hidden tests, making this one of the most rigorous assessments of AI research capabilities available.venturebeat+5

Claude Fable 5.1: General Availability with Production-Grade Safeguards

Claude Fable 5.1 is now generally available to all users through the Claude API (accessible as claude-fable-5-1), the Claude desktop app, and major cloud platforms including Amazon Web Services, Google Cloud, and Microsoft Azure. The model carries production-grade safeguards that filter requests related to cybersecurity exploits and life sciences research, ensuring responsible use while maintaining high performance across other domains.anthropic+4

On coding benchmarks, Fable 5.1 achieved 55.8% on Terminal-Bench 4.0, representing a 13% improvement over Fable 5's 42.0% score. The model also demonstrated strong performance on Humanity's Last Exam, scoring 60.9% without tool assistance. These results reflect substantial improvements in the model's ability to handle complex, multi-step programming tasks and extended reasoning challenges that require sustained attention over long periods.itmedia.co+4

Claude Mythos 5.1: Restricted Access for Cybersecurity and Life Sciences Professionals

Alongside Fable 5.1, Anthropic introduced Claude Mythos 5.1, a variant of the same underlying model with more permissive safeguards designed specifically for vetted professionals working in cybersecurity and life sciences. Mythos 5.1 is identical to Fable 5.1 in terms of core capabilities but offers reduced restrictions for approved users whose work is affected by the cybersecurity and life sciences safeguards.anthropic+4

Mythos 5.1 is available exclusively through two trusted access programs:anthropic+2

Cyber Verification Program (CVP): This program provides vetted cybersecurity defenders with access to models that have reduced cybersecurity safeguards, enabling deeper work in defensive security research and vulnerability discovery. The CVP currently offers access to certain Opus- and Sonnet-class models, with plans to extend Mythos-class model access in the near future.anthropic+3

Life Sciences Verification Program (LSVP): Developed in partnership with the U.S. government, this program enables life sciences professionals to use Claude Mythos 5.1 with safeguards specifically configured for professional research and development activities. All other safeguards remain in place, ensuring responsible use while allowing legitimate scientific work to proceed without unnecessary restrictions.anthropic+3

Currently, Mythos 5.1 is limited to a select group of U.S. organizations, though Anthropic is coordinating with the U.S. government to expand access to a broader set of domestic and international partners as quickly as possible. On benchmarks with its more permissive safety settings, Mythos 5.1 scored 60.9% on Terminal-Bench 4.0, outperforming Fable 5.1's 55.8%.itmedia.co+4

Dramatic Cost Reductions: 75% Cheaper Cache Reads, Up to 45% Lower Workload Costs

One of the most significant aspects of the Fable 5.1 launch is Anthropic's aggressive pricing strategy, which dramatically reduces costs for developers and enterprises. While the base token pricing remains unchanged at $10 per million input tokens and $50 per million output tokens, Anthropic has slashed the price of cache reads by 75%, from $1.00 to just $0.25 per million tokens.venturebeat+3

Cache reads occur when the model reuses context it has already processed, a common pattern in agentic workflows where the same background information is referenced repeatedly. This pricing change has outsized effects on total costs for many use cases. Anthropic estimates that Fable 5.1 costs approximately 25% less than Fable 5 for typical workloads billed by token. For highly agentic tasks, where cached context makes up a larger portion of the bill, the estimated cost reduction reaches roughly 45%.venturebeat+4

These cost improvements make Fable 5.1 significantly more competitive for production deployments, particularly for applications that involve extended conversations, multi-step reasoning, or repeated references to the same context.venturebeat+2

Early User Feedback: Natural Writing, Fewer Interventions, Strong Long-Task Performance

Early users and developers who have tested Fable 5.1 have praised several key improvements over previous models. Common themes in feedback include:venturebeat+3

More Natural Writing Style: Users report that Fable 5.1 produces text that feels more human-like and less formulaic than earlier Claude versions, with improved tone, flow, and stylistic variation.intelligentliving+1

Fewer Interventions in Agentic Workflows: The model demonstrates greater autonomy and reliability when executing multi-step tasks, requiring less human oversight and correction compared to Fable 5 and Opus 5.venturebeat+2

Strength in Long-Horizon Tasks: Fable 5.1 excels at extended projects that require sustained attention over hours or days, such as building complete applications from scratch, conducting comprehensive research analyses, or developing complex codebases. Early demonstrations include the model building full games, designing protein binders with near-50% success rates, optimizing GPU inference speeds for open-source models by up to 2.5x, and creating high-precision maps of Venus using NASA radar data from over three decades ago.venturebeat+3

Improved Coding Capabilities: Developers highlight the model's enhanced ability to understand complex codebases, debug intricate issues, and generate production-quality code across multiple programming languages and frameworks.itmedia.co+2

Availability and Usage Limits Reset

Claude Fable 5.1 is available immediately across all Anthropic platforms, including the Claude API, Claude desktop applications for macOS and Windows, and partner cloud services. With the launch, Anthropic has also reset usage limits for all users, providing fresh allocations for the new model tier.anthropic+4

Claude Mythos 5.1 remains restricted to participants in the Cyber Verification Program and Life Sciences Verification Program, with access expanding gradually as Anthropic completes vetting procedures for additional organizations. The company emphasizes that both models maintain robust safety measures, with Mythos 5.1's reduced safeguards applying only to narrowly defined cybersecurity and life sciences use cases under strict oversight.anthropic+5

System Card Discloses Risks: Exploit Success Rates and Monitoring Challenges

In its accompanying System Card documentation, Anthropic transparently disclosed certain risks associated with Fable 5.1's advanced capabilities. The company reported that the model's improved reasoning abilities correlate with increased success rates on certain exploit attempts and noted challenges in monitoring model behavior for potentially harmful outputs. These disclosures reflect Anthropic's commitment to responsible AI development and provide users with important context for understanding the model's capabilities and limitations.anthropic+2

The company also highlighted that biology-related safeguards for Fable 5.1 and Fable 5 now fire 85% less often for benign requests related to elementary biology and medical questions compared to the original Fable 5 launch, reducing false positives while maintaining protection against genuinely risky queries. However, queries related to advanced research and development in the life sciences are still directed to Opus models or require Mythos 5.1 access through the LSVP.anthropic+2

Competitive Landscape: Anthropic vs. OpenAI, Google, and Others

The Fable 5.1 launch comes amid intensifying competition in the frontier AI model market. On the same day as Anthropic's announcement, OpenAI revealed that its Astra model series had crossed critical thresholds in cybersecurity capabilities, while Google continues to advance its Gemini models with improved multimodal reasoning. Fable 5.1's strong benchmark performance, particularly on Terminal-Bench-Science 0.1 where it more than doubles Fable 5's score, positions Anthropic competitively in this rapidly evolving landscape.venturebeat+3

The combination of superior performance, significant cost reductions, and specialized access programs for high-risk domains represents a strategic approach that balances capability advancement with responsible deployment. As enterprises and developers evaluate which models to integrate into their workflows, Fable 5.1's 25-45% cost savings alongside improved performance may prove decisive in tipping adoption toward Anthropic's platform.venturebeat+5

The Future of Agentic AI: From Answering Questions to Solving Problems

Fable 5.1's performance on benchmarks like Terminal-Bench-Science 0.1 illustrates a broader trend in AI development: the shift from models that simply answer questions to systems that can actively solve complex, multi-step problems. The ability to work autonomously for extended periods, reference cached context efficiently, and produce high-quality outputs across diverse domains positions Fable 5.1 as a tool for genuine knowledge work rather than just information retrieval.venturebeat+3

As Anthropic continues to expand access to Mythos 5.1 for vetted professionals and refine safeguards based on real-world usage, the company aims to enable breakthrough research and development while maintaining appropriate guardrails. For the broader AI community, Fable 5.1 sets a new benchmark for what's possible in agentic coding and scientific research, challenging competitors to match both its capabilities and its commitment to responsible deployment