Claude Pricing 2026: Fable 5.1, Opus 5 & Sonnet 5 — What Businesses Actually Pay
The short answer. Claude API pricing in 2026 ranges from $2 per 1M input tokens (Sonnet 5) to $10 per 1M input tokens (Fable 5.1). Anthropic's flagship Claude Fable 5.1 — released Sept 1, 2026 with a 1M-token context window and 128K max output — costs $10 per 1M input and $50 per 1M output ($0.25 cache read, down 75% from Fable 5's $1.00); the value-tier Claude Opus 5 costs exactly half — $5/$25; and Claude Sonnet 5 is $2/$10, made permanent on Aug 10, 2026. But list prices are not what businesses actually run: Ramp's August AI Index shows Fable 5 (the predecessor) — Anthropic's most expensive model by far — drew only ~6% of Anthropic business tokens and 11.4% of model-attributed Anthropic spend in July 2026, while cheaper Opus 5 has already passed it in enterprise spending. For most business workloads, the best-value Claude is Opus 5 or Sonnet 5, not the flagship.
Claude model pricing at a glance
Verified API rates (Anthropic official pricing, checked September 1, 2026):
| Model | Input / 1M | Output / 1M | Cache read / 1M | Launch | Status |
|---|---|---|---|---|---|
| Claude Fable 5.1 | $10.00 | $50.00 | $0.25 | Sept 1, 2026 | GA flagship (claude-fable-5-1); 1M-token context, 128K max output; cache reads −75% vs Fable 5 |
| Claude Mythos 5.1 | $10.00 | $50.00 | $0.25 | Sept 1, 2026 | Trusted access only (CVP/LSVP) — NOT GA; same base model as Fable 5.1, cyber/bio safeguards |
| Claude Opus 5 | $5.00 | $25.00 | — | Late Jul 2026 | Value frontier; exactly half of Fable 5.1 |
| Claude Sonnet 5 | $2.00 | $10.00 | $0.20 | Aug 10, 2026 | Permanent — Sept 1, 2026 increase cancelled |
Cross-vendor reference: OpenAI's GPT-5.6 Sol is on a promotional $4/$20 per 1M through Nov 21, 2026 (base $5/$30) — the direct comparison in the adoption table below.
Fable 5.1 / Fable 5 vs Opus 5 vs GPT-5.6 Sol: Ramp adoption comparison
Ramp's August 2026 AI Index (published Aug 12, 2026) tracks what businesses connected to Ramp's AI cost tracker actually purchased in July 2026 — list price is not the same as what businesses run. Note: the July data measured Fable 5, the predecessor; Anthropic shipped Fable 5.1 (same $10/$50 list, cache reads cut from $1.00 to $0.25) on Sept 1, 2026. The flagship is the most expensive Anthropic model by far, yet it accounts for a small minority of usage:
| Model | API price (in/out per 1M) | Share of provider tokens | Share of provider spend | Launch | What the July 2026 data shows |
|---|---|---|---|---|---|
| Claude Fable 5 | $10 / $50 | ~6% of Anthropic business tokens | 11.4% of Anthropic model-attributed spend | Jun 9, 2026 | Anthropic's most expensive model by far, but a minority of usage; generated ~75% as much model-attributed spend as GPT-5.6 Sol in July (superseded by Fable 5.1, Sept 1, 2026) |
| Claude Opus 5 | $5 / $25 | — | Passed Fable 5 in enterprise spend (FT-cited Ramp data) | Late Jul 2026 | Cheaper model pulling ahead of the flagship in enterprise spend after its late-July debut — price and task sufficiency over maximum capability |
| GPT-5.6 Sol (OpenAI) | $4 / $20 promo (base $5/$30) | ~25% of OpenAI tokens | 23% of OpenAI spend | 2026 | Roughly half Fable 5's price; the volume leader across both vendors — Fable 5 is less popular with businesses than Sol overall |
- Attribution caveat on the Opus overtake: "Opus 5 already passed Fable 5 in enterprise spend" is reported from Ramp data relayed by the Financial Times (via Superpower Daily / CIO). Ramp's own AI Index does not state the overtake, and WinBuzzer cautions Opus arrived only near the end of the measured month, so it cannot explain the full July pattern. Treat it as the direction of travel, not a settled July stat.
- Ramp's own caveat: the model-level sample "skews slightly more tech-y than our typical AI Index sample. So actual Fable adoption is likely even lower than what we have estimated here."
- Broader July adoption: 43.5% of US businesses on Ramp's tracker paid Anthropic vs 39.7% for OpenAI — Claude as a platform is the most widely adopted, even as its flagship lags on spend.
For the same price-vs-adoption exercise applied to coding agents, see our AI coding agent pricing comparison.
Is Claude too expensive? What the Ramp data says
The honest answer: the flagship is, which is why businesses route around it. Ramp's data shows Fable 5 — at $10/$50 per 1M, roughly 2x the price of GPT-5.6 Sol and 5x Sonnet 5 on input — captured just 6% of Anthropic business tokens and 11.4% of model-attributed spend in July. One caveat: that measured the predecessor. On Sept 1, 2026 Anthropic shipped Fable 5.1 at the same $10/$50 list but with cache reads cut 75% (to $0.25/1M) — cheaper for exactly the cache-heavy agent workloads where the flagship premium hurt most. The cheaper Opus 5 moving ahead of the flagship in enterprise spending reinforces the point: businesses do not necessarily want maximum capability all the time, and price wins volume.
That said, "Claude is too expensive" is not the same as "Claude is losing." In July, more US businesses paid Anthropic (43.5%) than OpenAI (39.7%) — Claude's platform-level adoption is the strongest of any vendor. The expensive part is the flagship tier, and the market is voting with tokens for Opus 5 and Sonnet 5 instead.
Agency takeaway: don't default client quotes to Fable 5.1 list rates. Price retainers against a realistic model mix — Opus 5 / Sonnet 5 / GPT-5.6 Sol capture the volume, and Fable 5.1 is a premium line for quality-critical work (with Fable 5.1's $0.25/1M cache reads, quote it for cache-heavy agent stacks rather than blanket premium). You can model that mix directly in the AI Agency Pricing Calculator, which now includes Fable 5.1, Opus 5, and GPT-5.6 Sol rows.
Best Claude model for business
- Claude Opus 5 ($5/$25) — the enterprise value frontier. Near-flagship quality at exactly half of Fable 5.1's price, and per FT-cited Ramp data the model enterprise spend is moving to. Published per-task reference ≈ $2.34 per task (Artificial Analysis, Aug 2026) — a ~6x spread vs cheap small models on the same class of work.
- Claude Sonnet 5 ($2/$10) — the cost-efficient production workhorse. Input at open-weight parity (Qwen 3.8 Max $2), permanent pricing with no scheduled increase, cache reads at $0.20/1M. The default for high-volume production workloads.
- Claude Fable 5.1 ($10/$50) — maximum capability (1M-token context, 128K max output, released Sept 1, 2026), for quality-critical work where the premium (roughly 2x GPT-5.6 Sol) is justified. Watch the price and the fact that most businesses didn't run the predecessor; Fable 5.1's $0.25/1M cache reads and EFS/ZDR options address the two big adoption blockers.
Rule of thumb: route by task, benchmark on your own workload, and re-run quarterly — 2026 pricing moves weekly (GPT-5.6 Sol Aug 21, DeepSeek Aug 16, Gemini 3.7 Flash Aug 13, Grok 4.6 Aug 12, Claude Sonnet 5 Aug 10).
Fable 5.1's data-retention changes: EFS and ZDR
There was a governance catch on the flagship — and Fable 5.1 changes it. Under Anthropic's June 9, 2026 policy, a mandatory 30-day data-retention requirement applied to Fable 5 and Mythos 5 traffic. With Fable 5.1 (Sept 1, 2026), Anthropic introduced Enterprise Frontier Safeguards (EFS): customer data is stored in cloud infrastructure controlled entirely by the customer — complete privacy, the same as a zero-data-retention policy.
- Until EFS is available, eligible customers can use Fable 5.1 with zero data retention — the blanket 'no ZDR on the flagship' rule no longer applies to 5.1.
- EFS keeps ZDR-equivalent privacy while remaining state-of-the-art at preventing adversarial use. It was developed with 100+ customers across financial services, healthcare, manufacturing, telecom, law, retail, and the public sector, and with AWS, Google Cloud, and Microsoft Azure; it is supported on Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Google's Agent Platform, and Microsoft Foundry. EFS rolls out in phases starting fall 2026.
- This matters for enterprises with strict data-handling requirements: the old Fable 5 retention barrier (DigitalApplied documents Microsoft limiting internal Fable 5 use) is being removed for Fable 5.1 — confirm EFS/ZDR eligibility for your account before routing sensitive traffic.
- Pricing is otherwise unchanged: $10 per 1M input / $50 per 1M output, with cache reads cut to $0.25/1M.
Developer sentiment: "Claude Code treats you like a child getting lootboxes"
The pricing pressure is showing up in developer sentiment too. On Aug 24, 2026, Pieter Levels (levels.io) posted: "Claude Code treats you like a child getting lootboxes so I'll probably switch to xAI" — and it was not a one-off: his August archive also includes "I'm deleting Claude" (Aug 8), a prompt-workaround post (Aug 7), and a complaint about verbose output (Aug 3).
Anecdotes are not data, but the sentiment matches Ramp's signal: when the flagship's price and usage limits exceed what the work is worth, developers and businesses route to cheaper alternatives — Opus 5, GPT-5.6 Sol, or xAI.
Model Fable 5.1, Opus 5, and GPT-5.6 Sol side by side
Open the AI Agency Pricing Calculator →The calculator includes Claude Fable 5.1 ($10/$50), Claude Opus 5 ($5/$25), and GPT-5.6 Sol ($4/$20 promo) strategies with the July adoption data.
Frequently asked questions
What is Claude Fable 5.1 API pricing?
Claude Fable 5.1, Anthropic's flagship released Sept 1, 2026, costs $10 per 1M input tokens and $50 per 1M output tokens, with cache reads at $0.25 per 1M (down 75% from Fable 5's $1.00; 0.025x base input price). It carries a 1M-token context window and 128K max output (API id claude-fable-5-1). Claude Mythos 5.1 is the same underlying model with cyber/bio-tuned safeguards, priced identically but available only through trusted-access programs — not GA.
Is Claude too expensive for businesses?
The flagship is, but most businesses don't run the flagship. Ramp's August 2026 AI Index measured July 2026 business purchases: Claude Fable 5 drew only ~6% of Anthropic business tokens and 11.4% of model-attributed Anthropic spend despite being the most expensive model by far, while cheaper Claude Opus 5 ($5/$25 per 1M) has already passed it in enterprise spend per FT-cited Ramp data. At the same time, 43.5% of US businesses on Ramp's tracker paid Anthropic in July vs 39.7% for OpenAI — Claude overall is the most widely adopted vendor, but price-sensitive businesses are choosing Opus 5 and Sonnet 5 over Fable 5.
What is the best Claude model for business?
For most business workloads, Claude Opus 5 ($5/$25 per 1M, exactly half of Fable 5.1, launched late July 2026) is the best value — near-flagship quality at half the price, and the model enterprise spend is moving to per FT-cited Ramp data. Claude Sonnet 5 ($2/$10 per 1M, permanent since Aug 10, 2026) is the cost-efficient production workhorse. Claude Fable 5.1 ($10/$50, 1M context, 128K max output) is worth its premium only for quality-critical work where maximum capability justifies roughly 2x the price of GPT-5.6 Sol.
How does Claude pricing compare to GPT-5.6 Sol?
Claude Fable 5.1 is roughly twice as expensive as OpenAI's GPT-5.6 Sol — $10/$50 per 1M vs Sol's $4/$20 promotional rate (base $5/$30) through Nov 21, 2026. Ramp's August 2026 AI Index shows GPT-5.6 Sol holds ~25% of OpenAI tokens and 23% of OpenAI spend, while Fable 5 (the predecessor, measured in July) generated only ~75% as much model-attributed spend as Sol — the cheaper model wins the volume.
Does Claude Fable 5.1 require 30-day data retention?
Fable 5.1 changes the picture. The June 9, 2026 30-day retention policy applied to Fable 5 and Mythos 5. With Fable 5.1, Anthropic introduced Enterprise Frontier Safeguards (EFS): customer data lives in cloud infrastructure controlled entirely by the customer — complete privacy, the same as a zero-data-retention policy. EFS rolls out to enterprise customers in phases starting fall 2026; until EFS is available, eligible customers can use Fable 5.1 with zero data retention. Organizations should still confirm EFS/ZDR eligibility for their account before routing sensitive traffic.
Sources
- Ramp AI Index, August 2026 — "Cracks in the AI thesis" (Ara Kharazian, published Aug 12, 2026): ramp.com/data/ai-index-august-2026
- WinBuzzer (Aug 26, 2026) — "Businesses Avoid Spending For Anthropic's Expensive Fable 5 Model": winbuzzer.com
- BreezyScroll (Aug 24, 2026) — "Anthropic's Top AI Model Struggles To Win Business Users": breezyscroll.com
- Superpower Daily (Aug 24, 2026) — "Fable 5 Has 11% of Enterprise Spend as Opus 5 Pulls Ahead" (FT-cited Ramp data): superpowerdaily.com
- levels.io August 2026 archive — Pieter Levels, "Claude Code treats you like a child getting lootboxes so I'll probably switch to xAI" (Aug 24, 2026): levels.io/2026/august
- ByteChat (Jul 5, 2026) — "Claude Fable 5 30-Day Data Retention Rule": bytechat.io
- DigitalApplied (Jul 2, 2026) — "Fable 5's 30-Day Retention & ZDR": digitalapplied.com
- Anthropic (Sept 1, 2026) — "Claude Fable 5.1 and Claude Mythos 5.1" (release announcement: pricing, EFS/ZDR, safeguards): anthropic.com/claude-fable-and-mythos-5-1
- Anthropic official pricing (verified Sept 1, 2026; Fable 5.1/Mythos 5.1 $0.25 per 1M cache reads): platform.claude.com/docs/en/about-claude/pricing
Accuracy note: All Ramp figures verified 2026-08-26 against the Ramp primary report and corroborating coverage (research brief t_06f1cfb2, 7 sources, 31 evidence quotes); the July data measured Claude Fable 5, superseded by Fable 5.1 on Sept 1, 2026. The "Opus 5 passed Fable 5 in enterprise spend" claim is attributed to FT-cited Ramp data (Superpower Daily / CIO) — Ramp's own AI Index does not state the overtake, and WinBuzzer cautions Opus arrived too late in the measured month to explain the full July pattern. Fable 5.1 / Mythos 5.1 specs (1M-token context, 128K max output, $10/$50 per 1M, $0.25/1M cache reads, Mythos 5.1 trusted-access-only) are sourced to Anthropic's Sept 1, 2026 announcement (anthropic.com/claude-fable-and-mythos-5-1) and the official platform pricing docs (verified Sept 1, 2026); the June 9, 2026 retention policy applied to Fable 5/Mythos 5 and is superseded for Fable 5.1 by EFS/ZDR as covered by ByteChat/DigitalApplied/Anthropic. GPT-5.6 Sol pricing is OpenAI's official promotional rate ($4/$20 per 1M, cached input $0.40, valid Aug 21 – Nov 21, 2026; base $5/$30). Per-1M rates verified Sept 1, 2026 and move frequently — re-verify before quoting. No hands-on model testing was performed for this page; all figures trace to named sources.