Fable 5.1 Cuts Agent Costs, Doubles Science Benchmark
💡Fable 5.1 targets persistent Agents with major benchmark gains and 75% cheaper cache reads.
⚡ 30-Second TL;DR
What Changed
Fable 5.1 retains a 1-million-token context window, 128K maximum output, and pricing of $10 per million input tokens and $50 per million output tokens.
Why It Matters
The update strengthens Anthropic’s position in high-value coding, research, and enterprise automation workflows. Lower cache-read costs may make persistent Agents more economically viable, although the model’s high output-token price still limits broad adoption.
What To Do Next
Run a representative long-horizon coding Agent on the Fable 5.1 API and measure cache-hit rates, total tokens, latency, and cost against Fable 5.
Key Points
- •Fable 5.1 retains a 1-million-token context window, 128K maximum output, and pricing of $10 per million input tokens and $50 per million output tokens.
- •Cache-read pricing falls from $1 to $0.25 per million tokens, with estimated total workload savings of about 25% and up to 45% for complex Agent coding tasks.
- •Terminal-Bench-Science improves from 24.7% to 52.6%, while AutomationBench rises from 17.1% to 31.4%.
- •Early tests show stronger autonomous execution, including a 38-hour uninterrupted experiment run by Ramp.
🧠 Deep Insight
Background and context from public sources — not the original article. 10 sources cited.
🔑 Enhanced Key Takeaways
- •Anthropic introduced Enterprise Frontier Safeguards (EFS), a security architecture allowing organizations to retain monitoring data within their own controlled cloud infrastructure.
- •Mythos 5.1 features specialized safeguard layers that reduce false-positive blocking by 60% in cybersecurity vulnerability discovery tasks compared to previous iterations.
- •Fable 5.1 outperforms the GPT-5.6 Sol model (22.4%) and the previous Opus 5 (29.0%) on the Terminal-Bench-Science 0.1 benchmark.
- •The model incorporates 'adaptive thinking' enabled by default, which contributes to the observed reduction in shortcut-taking behavior during autonomous testing.
- •Fable 5.1 is simultaneously deployed across major cloud ecosystems including Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry.
📊 Competitor Analysis▸ Show
| Feature | Fable 5.1 | GPT-5.6 Sol | Opus 5 |
|---|---|---|---|
| Terminal-Bench-Science | 52.6% | 22.4% | 29.0% |
| Primary Focus | Long-running Agents | General Purpose | Research/Reasoning |
| Cache Read Pricing | $0.25/M tokens | N/A | $1.00/M tokens |
🛠️ Technical Deep Dive
- Architecture: Utilizes a dual-model deployment strategy where Fable 5.1 and Mythos 5.1 share a common base model but diverge at the safeguard layer.
- Security: Implements Enterprise Frontier Safeguards (EFS) to decouple monitoring data from Anthropic's internal systems, enabling customer-controlled data residency.
- Agentic Logic: Enhanced reasoning capabilities specifically tuned to prevent 'test-skipping' behaviors, forcing the model to address root causes in software debugging.
- Optimization: Cache-read efficiency improvements achieved through architectural refinements in the attention mechanism's interaction with the prompt cache.
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (10)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: 虎嗅 ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
Weekly AI briefing
One email a week. Unsubscribe anytime.