Anthropic has announced Claude Haiku 5.5, a faster and lower-cost model intended to replace the older Haiku 4.5. The previous Haiku was released nearly a year ago and was priced at $1 per million input tokens and $5 per million output tokens. Haiku 5.5 matches OpenAI’s GPT-6 Luna pricing up to 100,000 tokens: $0.10/$0.50 (input/output). Beyond 100,000 tokens, Haiku 5.5’s price increases fivefold to $0.50/$2.50.
For comparison, GPT-6 Luna applies its price increase at 272,000 tokens and then charges $0.20/$0.75.
Tokenization and a hidden cost increase
Haiku 5.5 uses a new, less generous tokenizer. Using the Claude Token Counter tool mentioned by the author, the same long prompt consumes roughly 1.25× as many tokens on Haiku 5.5 as it did on Haiku 4.5, which effectively raises the cost for long prompts beyond the nominal per-token price.
If your workloads fit within 100,000 tokens, Haiku 5.5 is priced the same as Luna and reportedly achieves higher benchmark scores. For usage above 100,000 tokens, Luna appears to be the more cost-effective option.
Practical test and results
The author updated the llm-anthropic plugin (the latest llm-anthropic release removed the need to ship a new plugin for every model) and tested the model with these commands:
- llm install -U llm-anthropic
- llm anthropic refresh
- llm -m claude-haiku-5.5 "Generate an SVG of a pelican riding a bicycle" -o thinking_effort low
They generated pelican SVGs at low, medium, high, xhigh and max reasoning effort. Haiku 5.5 does not allow disabling reasoning and defaults to medium. The low-effort pelican took 7 seconds and cost 0.0936 cents. The max-effort pelican took 5 minutes 9 seconds and cost 3.3826 cents.
By comparison, a pelican generated a year earlier with Haiku 4.5 cost 0.7583 cents and produced a substantially worse drawing, according to the author.
Sonnet 5.5 cache costs and subscription API credits
Anthropic also announced that it has halved the cache-read cost for Sonnet 5.5. In addition, the company added monthly API credits for Max and Team subscribers on the Claude Platform:
- Max 5x users receive $100 in API credits per month.
- Max 20x users receive $200 in API credits per month.
- Team subscribers receive up to $500 per month in pooled credits across their users.
Subscribers can claim these credits under Settings -> Billing by selecting the API organization that should receive the monthly credits. The credits match the subscription cost and do not roll over — unused credits are forfeited. Anthropic also allows disabling auto-reload; if auto-reload is turned off, API requests will stop when the balance is exhausted.
Why this matters
Monthly API credits make it significantly easier for subscribers to use Anthropic’s API because the credits cover the subscription cost. OpenAI still permits Codex subscriptions to be used for personal API access, which can be more economical for heavy API users; Anthropic’s credit program reduces that advantage but does not eliminate it.
In summary, Claude Haiku 5.5 offers competitive pricing and improved performance for sessions up to 100,000 tokens, while longer sessions may favor GPT-6 Luna. The Sonnet cost reduction and subscriber credits further incentivize Anthropic platform usage.
