Key Points:
- Anthropic launched Claude Fable 5.1 and Mythos 5.1, featuring faster reasoning, improved coding skills, and reduced operating costs.
- A 75% price cut on cache reads reduces token costs by approximately 25% on standard tasks and up to 45% on agentic workflows.
- Fable 5.1 scored 55.8% on the Terminal-Bench 4.0 coding benchmark, outperforming Fable 5 and Opus 5.
- Refined cybersecurity safeguards block 60% fewer false positives, allowing engineers to discover software vulnerabilities without refusal errors.
Anthropic released Claude Fable 5.1 and Claude Mythos 5.1, upgrading its flagship artificial intelligence models with faster execution, improved reasoning, and significantly lower operating costs. Fable 5.1 arrives with general availability for enterprise developers, while Mythos 5.1 serves specialized research teams through trusted access programs. The rollout addresses customer feedback regarding token pricing, software coding accuracy, and rigid safety filters that previously limited the utility of earlier frontier models.
A major breakthrough in Fable 5.1 centers on cost efficiency for large-scale software engineering. The new model delivers an estimated 25% cost reduction for standard token-billed workloads compared to the original Fable 5. Anthropic achieved these cost savings by slashing prompt cache read prices by 75%, cutting the cost of accessing stored context to one-quarter of its previous level. For complex agentic workflows that repeatedly reference massive codebases and technical documentation, overall project savings can reach up to 45%.
The model sets new performance benchmarks across automated coding and terminal operations. On the Terminal-Bench 4.0 evaluation, Fable 5.1 achieved a score of 55.8%, outperforming the 42.0% recorded by Fable 5 and the 52.3% posted by Opus 5. The unrestricted Mythos 5.1 model achieved 60.9% on the same benchmark, demonstrating unmatched capability in diagnosing complex system architectures, resolving software errors, and orchestrating multi-file pull requests.
In addition to coding capabilities, Fable 5.1 elevates performance in structured knowledge work, financial modeling, and scientific research. On the GDPval-AA benchmark for enterprise knowledge tasks, Fable 5.1 achieved a score of 1,853, surpassing Opus 5 at 1,824 and Fable 5 at 1,723. The model processes long-horizon assignments across complex spreadsheets, legal briefs, and presentation decks with greater contextual coherence and fewer hallucinations.
Anthropic redesigned its safety architecture to resolve user frustrations with false positives. Previous safety classifiers frequently refused benign programming prompts, mistaking standard software debugging and vulnerability scanning for cyberattacks. The updated guardrails in Fable 5.1 reduce false positive refusals in cybersecurity tasks by 60%. As a result, security researchers can use the model to uncover software flaws and audit legacy code without encountering constant refusal blocks.
For enterprise customers handling proprietary codebases and sensitive consumer data, Anthropic introduced Enterprise Frontier Safeguards (EFS). This privacy framework allows organizations to run frontier AI models with zero data retention by storing customer information exclusively on client-controlled cloud infrastructure. Scheduled for a phased rollout, EFS gives enterprise clients full data sovereignty while ensuring compliance with stringent corporate governance and regional data privacy standards.
The specialized Mythos 5.1 tier introduces advanced capabilities for high-stakes scientific fields. Developed in close coordination with United States government agencies, Mythos 5.1 includes specialized safeguards tailored for biological research, genomics, and defensive cybersecurity operations. Qualified laboratories and academic institutions can apply for vetted enrollment to accelerate therapeutic discovery, vaccine modeling, and automated defense networks.
Pricing for standard input and output tokens remains stable at $10 per million input tokens and $50 per million output tokens. By maintaining baseline pricing while drastically reducing cache read expenses, Anthropic encourages developers to build long-running autonomous agents that maintain deep persistent memory across multi-turn sessions. This economic model makes high-end reasoning financially viable for production environments running around the clock.
The release intensifies competition across the frontier artificial intelligence market. As rivals OpenAI and Google rapidly iterate their flagship models, Anthropic is focusing on practical developer utility, superior code synthesis, and predictable enterprise costs. By pairing high benchmark scores with lower operational overhead, Anthropic aims to capture a dominant share of the high-margin corporate software engineering market.
Claude Fable 5.1 demonstrates that artificial intelligence development is entering an era where practical reliability, cost-efficiency, and user-centric guardrails matter just as much as raw computational power. With its balanced pricing structure, refined safety filters, and top-tier coding performance, Fable 5.1 equips engineering teams and corporate enterprises with an agile tool designed to accelerate complex digital innovation.
