SAT, AUGUST 01, 2026
Independent · In‑Depth · Practitioner‑Tested
✎ Large Language Models

Claude Sonnet 5 Intro Pricing Ends August 31 — 50% Cost Increase Incoming and a Tokenizer Trap to Know

Claude Sonnet 5 intro pricing ($2/$10/M) ends August 31, 2026 at 11:59 PM UTC. September 1: $3/$15/M — a 50% increase on every token. Second factor most teams miss: the new tokenizer adds up to 35% more tokens for same text (code and JSON heaviest). Combined: up to ~95% more per equivalent text unit at standard pricing. No extension announced. Front-load batch jobs, enable prompt caching, model the annual impact now.

By AIToolsRecap August 1, 2026 6 min read 26 views
Home Articles Large Language Models Claude AI Claude Sonnet 5 Intro Pricing Ends August 31 — ...

ACTION REQUIRED — 31 DAYS UNTIL AUGUST 31, 2026

Current price (through Aug 31): $2/M input · $10/M output
Price from September 1: $3/M input · $15/M output (+50% on every token)
New tokenizer uplift: Same text = up to 1.35× more tokens (heaviest on code, JSON, non-English)
Combined impact at standard pricing: Up to ~95% more per equivalent text unit vs Sonnet 4.6 at standard price
No extension announced: Confirmed on Anthropic's pricing docs — hard deadline 11:59 PM UTC August 31
Bedrock and Vertex AI: Same deadline — Anthropic pricing passes through on both
Model string: claude-sonnet-5 (unchanged — no action needed for model selection)

The Two Cost Mechanics — And Why Most Teams Miss the Second One

According to Anthropic's official Sonnet 5 page, the introductory price of $2/$10/M ends August 31, 2026, after which standard pricing of $3/$15/M takes effect. That is the headline number most teams have seen. As Finout's FinOps analysis documents, there is a second cost mechanism most teams have not modelled: Sonnet 5 ships with a new tokenizer that maps the same text to approximately 1.0-1.35× more tokens than Sonnet 4.6 — heavier on code, structured data, and non-English text, lighter on natural language prose. This means the same text that cost you $2/M at intro pricing will cost up to $3 × 1.35 = $4.05/M-equivalent at standard pricing from September 1 — a combined increase of up to 100%+ per equivalent text unit, not the headline 50%.

As the official Sonnet 5 release notes state directly: "Because the new tokenizer produces approximately 30% more tokens for the same text, the cost of an equivalent request can differ from Claude Sonnet 4.6 even though per-token pricing is unchanged." Anthropic frames the model as "roughly cost-neutral" at intro pricing vs Sonnet 4.6 at its standard price — and that is accurate for the intro window. It is not accurate from September 1.

What This Means for Real Workloads

Monthly volumeNow (intro $2/M)Sept 1 ($3/M)Monthly increase
10M tokens$20$30+$10
100M tokens$200$300+$100
500M tokens (mid-tier)$1,000$1,500+$500/month
1B tokens$2,000$3,000+$1,000/month

Input token cost only. Output token increase is proportionally identical (+50%). Add tokenizer uplift (10-35% more tokens per equivalent text) on top of these figures for code-heavy or non-English workloads. Per FinOps LLM's modelling, a mid-tier workload of 500M input + 150M output tokens/month will see a $1,250 monthly increase that compounds to $15,000 annually.

What to Do Before August 31

1. Front-load batch processing jobs

Any large batch job — content generation, classification at scale, document processing — that can be run before August 31 should be. The $2/M vs $3/M input difference is real money at volume.

2. Measure your actual tokenizer uplift

The 30-35% tokenizer increase is an average. Run your actual workload through the Sonnet 5 tokenizer and measure what your specific content produces. Code, JSON, and non-English text are heavier. Conversational prose is lighter. Know your real uplift before September 1.

3. Evaluate prompt caching

Anthropic offers up to 90% cost savings with prompt caching on Sonnet 5. If your workload uses a stable system prompt or large context that repeats across requests, prompt caching significantly reduces the effective cost increase. Enable it now and measure the savings before the deadline.

4. Consider routing tier decisions after September 1

From September 1, Sonnet 5 at $3/M and Claude Haiku 4.5 (cheaper, faster) create a clear tier decision: Haiku 4.5 for high-volume simple tasks, Sonnet 5 for quality-requiring tasks, Opus 5 at $5/M for hard reasoning. Build routing logic now to shift appropriate workloads to Haiku 4.5 before the price step-up takes effect.

5. Budget the annual impact

A 50% increase in per-token cost plus 15-35% more tokens for the same content can compound to a 70-100% increase in annual AI infrastructure spend on Sonnet 5 workloads. Model this against your current monthly Anthropic bill before September 1 lands.

Sources: Anthropic Sonnet 5 pricing page · Official release notes (tokenizer) · FinOps LLM cost modelling · Finout FinOps analysis · Related: Sonnet 5 vs Opus 5 — decision framework → · Claude Opus 5 full review →

Tags
AnthropicClaude CodeAI NewsGenerative AI2026

Spot an inaccuracy?

We verify facts before publishing and correct errors promptly. If something in this article is wrong or outdated, let us know.

Report an error →