Anthropic's tokenizer change makes Opus 5.5 cost about 4% more per task than Opus 4.6, PromptUnit estimates
Anthropic's pricing page says Claude 4.7 and later models produce about 30% more tokens for the same text, so per-token cuts overstate upgrade savings. PromptUnit's recalculation assumes the full 30% on every model pair, so any one workload's result depends on how its own content tokenizes.
The Engineer · Build desk
PromptUnit estimates Opus 5.5 at 5.2 dollars per million input tokens once 1.3 times the tokens are counted, against 5 dollars for Opus 4.6 or 4.5. Anthropic's pricing page lists Opus 5.5 input at $4 per million tokens.
| Figure | Value | Note | Claim |
|---|---|---|---|
| Opus 5.5 at 1.3x the tokens, PromptUnit estimate | $5.2 per M input | 8 | |
| Opus 4.6 or 4.5 | $5 per M input | 8 | |
| Opus 5.5 list price | $4 per M input | 1 |
What happened
- Per million tokens, the list puts Haiku 5.5 at $0.10 for input and $0.50 for output. Sonnet 5.5 is $2 in and $10 out, Opus 5.5 is $4 in and $20 out, and Fable 5.1 is $10 in and $50 out. The Haiku rate applies only to prompts of up to 100,000 tokens.
- The post lists six costs the per-token table leaves out: the tokenizer change, Haiku 5.5 pricing for long prompts, US-only inference, fast mode, web search and code execution.
- Batch API pricing is half those rates. Reading from the cache costs 2.5% to 10% of the input price, and the share varies by model.
- Sonnet 5 launched at $2 and $10 as introductory pricing through August 31, 2026, and Anthropic has since made that the standard price.
- Haiku 4.5 is a legacy model whose retirement is listed as not sooner than October 15, 2026, with at least 60 days' notice promised before any retirement.
Why it matters
- What it costs cost Reasoning tokens bill at the output rate even when the thinking text is not returned, and on Opus 5.5 and Fable 5.1 they cannot be switched off, so the effort parameter is the control that remains.
- Who has to decide decision Forecasts that assumed the announced September 1 rise to $3 and $15 overstate Sonnet 5 token spend by 50%, so those budget lines can come down.
- Who has to decide decision Teams leaving a model on the old tokenizer need their own count before migrating; the post's route is a sample of real prompts through the token counting endpoint with each model's ID.
Cost per task is the per-token rate times the token count, and a newer tokenizer raises the count. The estimates below come from a DEV Community post by PromptUnit's founder, who says every price in it comes from Anthropic's documentation, read on October 14, 2026 [16].
On the full-30% assumption, Sonnet 4.6 to Sonnet 5.5 cuts the input rate from $3 to $2. The same task then costs 2 x 1.3 = 2.6 against 3, about 13% less [6]. Opus 4.6 or 4.5 to Opus 5.5 goes the other way: 4 x 1.3 = 5.2 against 5, about 4% more [8]. The Opus input rate falls 20% there [17], and a 20% cut covers a token increase of 25% and no more [18]. If a team's prompts grow by less than 25%, Opus 5.5 is cheaper per task.
Opus 4.7 and 4.8 users do not face this. Both sides use the newer tokenizer, so the post puts the same input at 20% less [9]. Output is another matter, because the post says its length still depends on the model and the effort setting [9].
Haiku 4.5 to Haiku 5.5 shows the largest swing: 0.10 x 1.3 = 0.13 against 1.00, about 87% less per task [7]. The post limits that figure to prompts up to 100,000 tokens [7]. If Haiku 5.5 counts that ceiling in its own tokens and the full 30% applies, a prompt of about 76,900 tokens on Haiku 4.5 already crosses it [19]. The post's figures for the rate above the ceiling and for the US-only inference premium were not in the material reviewed.
The post's first workload is a support chat: 500 input and 200 output tokens per call, 10,000 calls a day, with output counted to include thinking tokens [15]. On Haiku 5.5 the input costs $0.50 a day [15]. Output adds 2 million tokens at $0.50 per million, or $1.00, for $1.50 a day and $45 over a 30-day month [20].
What to watch
- Token counts from real Opus 4.6 traffic: an increase under 25% would put Opus 5.5 below Opus 4.6 per task.
- Anthropic's Haiku 4.5 retirement notice, which would start a 60-day migration clock onto the new tokenizer.
- The Haiku 5.5 rate above 100,000 tokens and the US-only inference premium on Anthropic's pricing page, read against a team's own prompt lengths.
Clarity's read
What the record supports and how the coverage leans. The claims behind it follow.
Reality
- Evidence55
- Adoption
- Not scored
- Hype gap+10
- Incentives
- Not scored
- Confidence55
Claim ledger
Ranked by verification strength, evidence, and original report placement.
- [1]
As of October 14, 2026, Anthropic's pricing page lists four current models: Haiku 5.5 at $0.10 input and $0.50 output per million tokens (for prompts up to 100,000 tokens), Sonnet 5.5 at $2 and $10, Opus 5.5 at $4 and $20, and Fable 5.1 at $10 and $50.
- [2]
The Batch API halves those prices, and a cache read costs between 2.5% and 10% of the input price, depending on the model.
- [3]
The post covers costs outside the per-token table: long prompts on Haiku 5.5, the tokenizer change, US-only inference, fast mode, web search and code execution.
- [4]
Anthropic's pricing page says Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text, and the exact increase depends on the content.
- [5]
The Haiku 5.5 overview says the same text counts as about 30% more tokens on Haiku 5.5 than on Haiku 4.5.
- [6]
Sonnet 4.6 to Sonnet 5.5: the per-token price drops by a third ($3 to $2 input, $15 to $10 output); with 30% more tokens the same task costs 2 x 1.3 = 2.6 against 3, about 13% less. The post calls this an estimate, not a measurement.
- [7]
Haiku 4.5 to Haiku 5.5 (prompts up to 100K tokens): 0.10 x 1.3 = 0.13 against 1.00, about 87% less per task. The post calls this an estimate.
- [8]
Opus 4.6 or 4.5 to Opus 5.5: 4 x 1.3 = 5.2 against 5 (dollars per million input tokens), about 4% more per task. The post calls this an estimate.
- [9]
Opus 4.7 or 4.8 to Opus 5.5: both sides use the newer tokenizer, so the same input costs 20% less. Output length still depends on the model and the effort setting.
- [10]
The post's estimates assume the full 30% token increase; the post says a team's own content may differ and advises running a sample of real prompts through the token counting endpoint with each model's ID to get its own number.
- [11]
Thinking tokens are billed as output tokens even when the thinking text is not returned. Thinking is on by default on Fable 5.1, Opus 5.5, Sonnet 5.5, Sonnet 5 and Haiku 5.5, and it cannot be turned off on Fable 5.1 and Opus 5.5.
- [12]
The effort parameter is the control: Opus 5.5 and Haiku 5.5 default to medium, the other models to high, and low spends the fewest tokens.
- [13]
Sonnet 5 launched at $2 and $10 as introductory pricing through August 31, 2026. Anthropic has made that the standard price, and the planned rise to $3 and $15 on September 1 will not happen.
- [14]
Haiku 4.5 is a legacy model, not deprecated. Anthropic's deprecations page lists its retirement as not sooner than October 15, 2026, and Anthropic gives at least 60 days' notice before retiring a publicly released model.
- [15]
Scenario 1 in the post is customer support chat: 500 input and 200 output tokens per call, 10,000 calls a day, 30-day months, with output tokens including any thinking tokens. On Haiku 5.5 the input is 500 x 10,000 = 5 million tokens x $0.10 = $0.50 a day.
- [16]
The post's author says they are the founder of PromptUnit, the post first appeared on the PromptUnit blog, and every Claude price in it comes from Anthropic's documentation, read on October 14, 2026.
- [17]
The Opus input rate falls 20% from Opus 4.6 or 4.5 to Opus 5.5 ($5 to $4).
- [18]
A 20% price cut offsets a token-count increase of at most 25%; Opus 5.5 is cheaper per task only if tokens grow by less than 25%.
- [19]
If the full 30% applies and the 100,000-token ceiling is counted in Haiku 5.5 tokens, a prompt of about 76,900 tokens on Haiku 4.5 reaches the ceiling.
- [20]
The Haiku 5.5 support-chat scenario costs $1.50 a day and $45 over 30 days: output is 200 x 10,000 = 2 million tokens x $0.50 = $1.00 a day, plus $0.50 of input.
- [21]
Budgets that assumed Sonnet 5 would rise to $3 and $15 on September 1 overstate its token spend by 50% against the $2 and $10 standard price.
Get this ledger inside ChatGPT. Ask about any of our stories and the answer brings its figure, each claim’s verdict and the sources. Free, no account.
Sources
1 independent publisher whose own reporting we read for this story.
- dev.toClaude API Pricing Per Million Tokens: All Models (Oct 2026)
1 article · October 16, 2026
Topics and entities
Follow any of these and your For You feed starts watching them — no settings page required.
Topics
- AI Cost Management and FinOpsFollow
- LLM TokenizationFollow
- Claude API pricingFollow