🚀 Released September 1, 2026 — 75% cache-read cut ($1.00→$0.25)

Claude Fable 5.1

Anthropic's new Mythos-class flagship. A new frontier for coding, knowledge work and long-horizon agents, with cache-read price cuts making it up to ~45% cheaper in practice.

$10.00
Input / 1M tokens
$50.00
Output / 1M tokens
$0.25
Cache read / 1M
1M
Context window
128K
Max output tokens
Last updated: | Sources: Anthropic announcement · Model overview · Pricing | All Pricing · Calculator · Mythos 5.1

Pricing (as of Sept 1, 2026 launch)

Verified against Anthropic's official pricing page (platform.claude.com). All prices in USD per 1M tokens.

TierInput5m Cache Write1h Cache WriteCache ReadOutputNotes
Standard$10.00$12.50$20.00$0.25$50.00Standard rate. Flat 1M context
Key pricing points:
Input, output, and cache writes are unchanged from Fable 5 ($10/$50, 5m writes $12.50, 1h writes $20).
Only cache reads dropped from $1.00 → $0.25 (a 75% cut) — now 2.5% of the base input price (vs 10% for other models).
* Anthropic estimates this yields ~25% lower cost on typical workloads and up to ~45% on highly agentic (tool- and context-heavy) workloads.

Primary sources: Anthropic announcement (Cost and availability) · platform.claude.com pricing (verified 2026-09-01). Batch API is 50% off (input $5 / output $25).

Model Specifications

GA Release
September 1, 2026
Knowledge Cutoff
June 2026
Context Window
1,000,000 tokens
Max Output
128,000 tokens
Input Modalities
Text + Image
Output Modalities
Text
Thinking
Adaptive (always on)
Default Effort
high
Tool Use
Supported
Model ID
claude-fable-5-1

Sources: Claude Platform model overview (verified 2026-09-01). All current models support text & image input, text output, multilingual, vision and tool use. Available on AWS Bedrock, Google Cloud (Vertex AI) and Microsoft Azure (Foundry).

Performance — a major leap over Fable 5

Benchmarks from Anthropic's announcement (evaluated with production safeguards enabled). Figures as published by each vendor.

BenchmarkFable 5.1Fable 5Claude Opus 5GPT-5.6 Sol
Agentic scientific research
Terminal-Bench-Science 0.1
52.6%24.7%29.0%22.4%
Agentic coding
Terminal-Bench 4.0
55.8%42.0%52.3%37.3%
Knowledge work
GDPval-AA v2
1853172318241711
Computer use
OSWorld 2.0 (partial)
77.9%72.9%75.4%
Multidisciplinary reasoning
Humanity's Last Exam (no tools)
60.9%57.8%56.6%
Business workflows
AutomationBench
31.4%17.1%26.9%19.6%
Agentic coding
CursorBench 3.2.0
73.4%70.5%70.0%67.2%

Terminal-Bench-Science more than doubling Fable 5's score (24.7% → 52.6%) is the headline change — the area Anthropic centers in this launch. Mythos 5.1 hits 60.9% on Terminal-Bench 4.0. HLE (with tools) is 65.0%.

Source: Anthropic announcement (verified 2026-09-01)

Fable 5.1 vs Mythos 5.1

Same model, different safeguards. The two share the same underlying model; only the strength of applied safeguards differs.

Claude Fable 5.1 — generally available to all users via the API, Claude.ai, Claude Code and Claude Cowork.
Claude Mythos 5.1 — trusted-access only. A version with relaxed safeguards in cybersecurity (CVP) and life sciences (LSVP) for vetted organizations. See the Mythos 5.1 page for details.

Competitor Flag­ship Price Comparison

All prices in USD per 1M tokens (standard rate). Cache-read handling is the key cost differentiator.

ModelProviderInputOutputCache ReadNotes
Claude Fable 5.1Anthropic$10.00$50.00$0.25New flagship (Mythos-class)
Claude Opus 5Anthropic$5.00$25.00$0.50For standard workloads. Half Fable 5.1's price
GPT-5.6 SolOpenAI$4.00$20.00$0.40Promo (thru 2026-11-21). Cheapest in absolute terms
Claude Fable 5Anthropic$10.00$50.00$1.00Previous gen. Fable 5.1 cuts only cache reads 75%

💡 How to read it: Fable 5.1 is for the hardest, longest-horizon agents. For price-per-performance, GPT-5.6 Sol ($4/$20) or Claude Opus 5 ($5/$25) are cheaper. Note that on agentic workloads dominated by cache reads, Fable 5.1's effective cost drops up to ~45%.

Sources: each provider's official pricing page (verified 2026-09-01). See full pricing comparison.

Ideal Use Cases

Fable 5.1 shines on long-running, high-difficulty reasoning and agentic work.

💻

Long-horizon agents

Multi-step autonomous runs. Cache-read price cut improves cost efficiency.

🐛

Hard coding & debugging

Root-causing rare crashes other models miss.

🔬

Scientific research & knowledge work

More than 2× Fable 5 on Terminal-Bench-Science. Complex analysis and hypothesis generation.

🧠

Hard reasoning

60.9% on Humanity's Last Exam (no tools). Tough multidisciplinary problems.

🖥️

Computer use

77.9% OSWorld 2.0 (partial). GUI-driven automation.

📊

Business workflows

31.4% AutomationBench (~1.8× Fable 5's 17.1%). Autonomous routine work.

Claude Family Selection Guide

ModelInputOutputWhen to Use
Claude Fable 5.1 ⬅$10.00$50.00Hardest, long-horizon agents and scientific research (max capability)
Claude Opus 5$5.00$25.00Standard coding, agents and enterprise work
Claude Sonnet 5$2.00$10.00Speed × intelligence balance. High-volume, daily work
Claude Haiku 4.5$1.00$5.00Fastest, high-volume, classification. Cost-first

💡 Strategy: Claude Opus 5 ($5/$25) is often sufficient. Escalate to Fable 5.1 when Opus 5 at higher effort still falls short. For cybersecurity / life-sciences work blocked by safeguards, Mythos 5.1 (application required).

Estimate Your Actual Costs

Compare token costs for Claude Fable 5.1 and all major models. Just enter your token counts.

🧮 Open Token Calculator →

📊 Full Pricing Comparison · 🛡️ Claude Mythos 5.1 · 🗓️ Release Timeline · 🔗 Anthropic Announcement

⚠️ Disclaimer

  • Pricing and specs verified against Anthropic's official pages (platform.claude.com / anthropic.com) on 2026-09-01. Subject to change without notice. Always check official pages before contracting.
  • This site is informational only. We are not affiliated with or endorsed by any provider.
  • Benchmark scores are as published by Anthropic. Your mileage may vary.