Anthropic released Claude Haiku 5.5 on October 7 2026, calling it the “cheapest, fastest, and most capable small model we’ve ever released” as the latest entry in its lightweight frontier tier.
Headline API pricing has been set at $0.10 per million input tokens and $0.50 per million output tokens for contexts up to 100,000 tokens, putting the new model at parity with OpenAI’s GPT-6 Luna at the head of the small-model bracket. Anthropic’s own comparison shows Claude Haiku 5.5 costs roughly 75% less to run than Haiku 4.5, an aggressive repositioning that resets the floor for sub-flagship inference economics across the AI sector.
Claude Haiku 5.5: The Numbers Behind the Story
Benchmarks back up the marketing claim. On the Artificial Analysis Intelligence Index the new release scores 43, up from 17 for Haiku 4.5 and above GPT-6 Luna’s 38. That 26-point lift between generations is one of the largest single-cycle jumps Artificial Analysis has recorded for a small model, and it moves Haiku from a clear tier-two player into contention for tier-one coding and reasoning workloads priced for scale.
Beyond raw intelligence, Claude Haiku 5.5 records a 40% hallucination rate across the same evaluation suite, the lowest of the three models tested against GPT-6 Luna and the prior Haiku generation. Anthropic also confirmed a 1 million token context window with a 128,000 token maximum output, matching the working envelope of its larger siblings and removing the truncated context penalty that previously pushed many agent developers toward Sonnet and Opus SKUs.
Why Claude Haiku 5.5 Matters for the Market
Perhaps the most consequential change for builders is the addition of an adjustable effort dial, the first time an effort parameter has shipped on a Haiku-class model. The default sits at medium, but operators can dial it down for ultra-cheap batch traffic or push it toward high for workloads where latency dominates cost. Pricing changes were not announced for the higher effort tier, suggesting Anthropic is using effort as a latency and quality lever rather than a price multiplier.
Distribution is unusually wide for a small-model launch. Claude Haiku 5.5 is live on the free Claude.ai plan on the same day as the API release, alongside paid Pro, Max, Team and Enterprise tiers. Anthropic is also routing the model through its first-party console, Amazon Bedrock and Google Vertex AI, a three-channel rollout that limits the risk of any single cloud partner throttling adoption.
How Claude Haiku 5.5 Was Discovered
The pricing math shapes that rollout in material ways. A developer moving a 50 million token daily workload from Haiku 4.5 to Claude Haiku 5.5 at list prices drops input costs from approximately $5.00 per million tokens at the older SKU to about $1.25 per million tokens blended, once the 75% reduction is applied. For high-volume agent fleets, that is the difference between a line item and a rounding error in monthly cloud bills.
Capability gains look equally stacked in the new model’s favor. The combination of a 43 AAII score, 1M context and a 40% hallucination rate places Claude Haiku 5.5 within touching distance of older Sonnet-class systems at a fraction of the per-token cost, undercutting the typical use case for legacy mid-tier models in customer support automation, code review pipelines and bulk document classification.
What Comes Next for Claude Haiku 5.5
Competitive positioning also sharpens with this release. By matching GPT-6 Luna’s input price exactly and undercutting its output by meaningful margins on context-heavy jobs, Anthropic has chosen a price-anchoring strategy that forces rivals to react on benchmark optics rather than dollar figures. The simultaneous free-tier availability on Claude.ai means consumer-grade users get exposure to the model before many paid-channel customers can wire up the API.
For enterprise buyers the calculus is straightforward. Claude Haiku 5.5 is positioned as a drop-in upgrade for existing Haiku 4.5 deployments where context length and reasoning quality have been the binding constraints, while remaining cheap enough to justify experiment budgets for teams previously priced out of frontier-grade inference. Anthropic will be watching attach rates across Bedrock and Vertex AI closely as the first real signal of whether its cheapest-fastest pitch converts into durable market share against GPT-6 Luna and the new wave of open-source small models.
Anthropic’s October roadmap now hinges on whether Claude Haiku 5.5 can translate its price-led launch into sustained developer mindshare, or whether OpenAI and the open-source community counter with their own cost-down cycles before the next quarterly review. Either way, Claude Haiku 5.5 has redrawn the small-model price floor on day one, and the rest of the tier will spend the coming weeks reacting to it.
The Claude Haiku 5.5 release also closes the gap between Anthropic’s pricing and the open-source frontier: by undercutting Haiku 4.5 by 75% while extending context fivefold, Anthropic has reset the small-model tier for the coming quarter.
Source: https://felloai.com/claude-haiku/