Anthropic completed its 5.5 generation in two steps: Claude Sonnet 5.5 on September 28 and Claude Haiku 5.5 on October 7, 2026. Sonnet 5.5 is an easy upgrade. Haiku 5.5 brings something new to Anthropic's small model: a price that depends on prompt length.
Claude Sonnet 5.5
- Model:
claude-sonnet-5-5, active, retirement not sooner than September 28, 2027. - Price: $2 input / $10 output per million tokens. Cache reads are $0.10 per million, half of Sonnet 5's $0.20; cache writes are $2.50.
- Speed: Anthropic says it generates output 30%+ faster than Sonnet 5.
- Thinking off is now
between_tools: if you run Sonnet with thinking disabled, switch to the newbetween_toolssetting, which keeps up-front thinking off, before moving to 5.5. - It is also Anthropic's recommended replacement for Sonnet 4.5, which retires on November 30, 2026.
Claude Haiku 5.5
- Model:
claude-haiku-5-5, available on the Claude Platform, AWS, Google Cloud and Microsoft Azure. - An adjustable effort setting, a first for Haiku.
- Pricing tiered by prompt length (per million tokens):
| Prompt up to 100K tokens | Prompt over 100K tokens | |
|---|---|---|
| Input | $0.10 | $0.50 |
| Output | $0.50 | $2.50 |
| Cache reads | $0.01 | $0.05 |
| Cache writes | $0.125 | $0.625 |
For comparison, Haiku 4.5 is $1.00 / $5.00 with no length tiers. Below 100K tokens, Haiku 5.5 costs a tenth of that.
Why the Haiku tiers matter
Most small-model traffic is short: classification, extraction, routing, short chat. That traffic lands in the cheap tier and gets a 90% price cut compared with Haiku 4.5. The tier only bites on long prompts, where every token of that request costs five times more.
Three things to do:
- Know your prompt-length distribution. If a lane's prompts cluster around 90-110K tokens, small changes in retrieved context flip requests between tiers. Trim or cap context to stay under the line, or budget for the higher tier.
- Don't estimate cost with one number. Dashboards and budgets that multiply tokens by a single price will under-count long prompts on tiered models.
- Route very long prompts deliberately. Above 100K tokens Haiku 5.5 is still cheap, but compare it with Sonnet 5.5 on quality for that size of input before treating it as the default.
NeuronGate angle
NeuronGate's catalog lists one base price per model. When the upstream reports what it actually charged for a request, NeuronGate settles at whichever is higher, that or the listed price times your tokens, so a long prompt is billed at its real tier rather than the short-prompt rate. See the model catalog for current prices.
FAQ
Is Haiku 5.5 cheaper than Haiku 4.5?
For prompts up to 100K tokens, yes: $0.10 / $0.50 versus $1.00 / $5.00 per million. Above 100K tokens it is $0.50 / $2.50, still half of Haiku 4.5.
What replaces Sonnet 4.5?
Anthropic recommends Claude Sonnet 5.5. Sonnet 4.5 (claude-sonnet-4-5-20250929) retires on November 30, 2026.
