Anthropic just rewrote the price floor for small AI models. Claude Haiku 5.5 landed October 7–8 with input token costs slashed by up to 90% versus its predecessor — and the reverberations are already hitting competitors where it hurts most: the budget inference market.

The Numbers Are Brutal for the Competition

For prompts under 100,000 tokens — which Anthropic says accounted for roughly 90% of all previous Haiku requests — the new model prices input tokens at $0.10 per million and output tokens at $0.50 per million. That's a clean 90% reduction from Haiku 4.5's $1.00 input / $5.00 output structure. Even for longer prompts above the 100K threshold, prices drop 50% from the prior version.

The benchmark story is equally aggressive. On the Terminal-Bench 4.0 coding test, Haiku 5.5 scores 39.2% — against a literal 0.0% for Haiku 4.5. The context window expands from 200,000 to one million tokens. And benchmarking platform Artificial Analysis ranks it the leading small-class model on its Intelligence Index with a score of 43, edging out GLM-5.3 Flash (42), Gemini 3.8 Flash (41), and OpenAI's GPT-6 Luna (38).

Anthropic is also sweetening the ecosystem play: Sonnet 5.5 cache-read prices are being cut in half to $0.10 per million tokens, a move the company says trims typical agent workload costs by roughly 20%. Monthly API credits are rolling out this week for Max and Team subscribers — $100/month for Max 5x users, $200 for Max 20x users, and up to $500 pooled for Team accounts.

Source: Alex Wang, applied AI, Rogo — "accurate enough that we'd trust it there and fast and cheap enough that we can run it a lot." link

The Efficiency Catch — and the Real Competitive Pressure

Not everything runs in Haiku 5.5's favor. Artificial Analysis flags significantly higher token consumption at elevated effort levels: at its highest setting, Haiku 5.5 burns through roughly 162,000 output tokens per task on the Intelligence Index — about three times the 50,000 GPT-6 Luna requires at comparable performance. On cost-per-task efficiency, GPT-6 Luna reaches a similar intelligence score at roughly a third of Haiku 5.5's cost.

Anthropic itself flags that Haiku 5.5 uses an updated tokenizer that consumes slightly more tokens per task than its predecessor. Real-world savings will likely be smaller than the raw per-token numbers suggest.

"Haiku 5.5 scores 43 at $0.21 per task, sitting just below the Pareto frontier. GPT-6 Luna (max) reaches a similar score at roughly a third of the cost."

- Artificial Analysis

The model is also available immediately across AWS, Google Cloud, and Microsoft Azure.

The Pricing War Isn't Slowing Down

Anthropic is building Haiku 5.5 explicitly as a subordinate component in multi-agent pipelines — handling bounded subtasks so that Sonnet and Opus can concentrate on heavier workloads. That architectural framing matters: it's a direct play for the agentic AI infrastructure buildout, where per-token economics at scale are make-or-break.

The lower-tier rates now match GPT-6 Luna's pricing directly, while Google's Gemini 3.5 Flash-Lite comes in at $0.30 per million input tokens and $2.50 per million output tokens. Anthropic just forced every major inference provider to recalculate their value proposition — and that pressure isn't easing anytime soon.

Follow @AITechWireIO for daily coverage.