Sonnet 5 Pricing Cliff Canceled: Anthropic Keeps the $2/$10 Rate
By Eric Bush · July 29, 2026 · 4 min read· Updated August 22, 2026
Update, August 22: the pricing cliff described in the original article has been canceled. Anthropic made Sonnet 5's $2/$10 rate permanent.
Anthropic originally announced a September 1 move from $2/$10 to $3/$15 per million tokens. Its official pricing documentation now says the increase will not occur. Teams can use $2/$10 as the standard direct-API rate while still checking cache, batch, regional, and partner-platform modifiers.
The Dollar Impact
For a medium project consuming ~50M input tokens and ~8M output tokens:
| Pricing | Input Cost | Output Cost | Total |
|---|---|---|---|
| Current standard ($2/$10) | $100 | $80 | $180 |
| Canceled increase ($3/$15) | $150 | $120 | $270 |
| Increase avoided | — | — | $90 |
The Hidden Tokenizer Tax
What most developers miss: Sonnet 5 uses a new tokenizer that encodes text ~30% less efficiently than Sonnet 4.6. The same prompt that was 10,000 tokens on Sonnet 4.6 becomes ~13,000 tokens on Sonnet 5.
At the permanent rate, this still comes out cheaper on input: 13K × $2/M = $0.026 versus 10K × $3/M = $0.030. Tokenizer behavior depends on the workload, so measure real request counts rather than applying one percentage to every repository.
Three Strategies After the Increase Was Canceled
1. Remove the artificial deadline. Do not front-load work merely to beat a price change that is no longer scheduled. Sequence migrations by engineering value and readiness.
2. Maximize prompt caching. Cache hits cost $0.20/M, or 10% of base input. Stable prompt prefixes can reduce repeated input cost substantially.
3. Evaluate Opus 5 for complex work. At $5/$25, Opus 5 costs 2.5 times Sonnet 5 on standard input and output. It can still be cheaper per completed project when higher capability prevents enough retries and review churn.
Model Comparison Using Current Standard Rates
| Model | Input/M | Output/M | Best For |
|---|---|---|---|
| Sonnet 5 | $2 | $10 | Balanced coding tasks |
| Opus 5 | $5 | $25 | Complex/agentic work (fewer retries) |
| Haiku 4.5 | $1 | $5 | Simple tasks, high-volume |
| GPT-5.6 Sol | $5 | $30 | Alternative frontier |
The $2/$10 window is no longer scheduled to close. Use our AI Cost Calculator with the current standard rate, and recheck Anthropic's official page before long-term commitments.
Want to calculate exact costs for your project?
Frequently Asked Questions
When does Claude Sonnet 5's $2/$10 pricing end?
Anthropic says $2/$10 is now the standard price and the previously scheduled September 1 increase will not occur.
Is Sonnet 5 cheaper than Sonnet 4.6?
Its standard token rates are lower at $2/$10 versus $3/$15. Measure tokenizer and retry behavior on your workload for the full comparison.
Should I switch to Opus 5 for complex tasks?
Test it where better first-pass success could offset its 2.5x standard token rate versus Sonnet 5.
Related Articles
Claude Sonnet 5 Launch: $2/$10 Promo Pricing Undercuts Opus 4.8 for Coding Agents
Anthropic released Claude Sonnet 5 at $2/M input and $10/M output, then made that rate permanent. We compare coding-agent cost with Opus 4.8.
Introductory Pricing Traps: How to Avoid the 'New Model' Discount Cliff
New AI models often launch at promo rates that expire, sometimes doubling on a set date. Learn to spot introductory pricing cliffs and keep them from wrecking your budget.
Anthropic GRAM Method: How 'Knowledge Switches' Could Change AI Model Pricing Tiers
Anthropic's GRAM adds removable knowledge modules to transformers. Could future AI pricing become modular — base model plus capability add-ons?