Anthropic released Claude Haiku 5.5 on October 7, 2026, cutting baseline pricing to compete directly with OpenAI while making reasoning steps mandatory for standard model requests. The release replaces Claude Haiku 4.5, which launched nearly a year earlier, and targets lightweight computing tasks with lowered entry rates that scale sharply once prompts cross a 100,000-token threshold.
For inputs up to 100,000 tokens, Haiku 5.5 matches the pricing of OpenAI’s GPT-6 Luna, which debuted last month. Anthropic charges $0.10 per million input tokens and $0.50 per million output tokens, a steep drop from Haiku 4.5’s rates of $1 per million input and $5 per million output tokens. Beyond 100,000 tokens, Haiku 5.5 rates rise fivefold to $0.50 per million input and $2.50 per million output tokens. By comparison, GPT-6 Luna maintains its baseline rates up to 272,000 tokens before rising to $0.20 per million input and $0.75 per million output tokens. Analysis by Simon Willison using his Claude Token Counter tool showed that Haiku 5.5 also incorporates a less generous tokenizer, requiring roughly 1.25 times more tokens to process the same long prompt than Haiku 4.5.
While Haiku 5.5 reports higher benchmark scores than GPT-6 Luna for workloads under 100,000 tokens, Anthropic altered how the model handles processing steps. Haiku 5.5 does not allow users to disable reasoning, setting its default thinking effort to medium across a configuration scale spanning low, medium, high, xhigh, and max. Simon Willison reported that testing via the llm-anthropic tool on an SVG generation prompt produced functional bicycle frames on all settings above low. The low-effort run completed in 7 seconds and cost 0.0936 cents, while the max-effort run took 5 minutes and 9 seconds and cost 3.3826 cents. By contrast, Haiku 4.5 ran without reasoning controls, struggled with the graphic prompt, and cost 0.7583 cents.
During the max-effort test, Haiku 5.5’s output trace explicitly recognized the task as the “classic pelican-on-bicycle SVG test,” prompting secondary evaluation on an armadillo prompt using the xhigh setting. Alongside the model launch, Anthropic halved the price of cache reads for Claude Sonnet 5.5.
Anthropic also introduced a monthly credit distribution on the Claude Platform that provides API allowances matching subscription fees. Under the program, Max 5x users receive $100 per month, Max 20x users receive $200, and Team subscribers receive up to $500 in pooled monthly credits. The credits do not roll over at month’s end. Users can disable auto-reload to halt requests when balances clear, though OpenAI still retains an advantage for high-volume developers by allowing personal API access through existing Codex subscriptions.