# Anthropic prices Claude Haiku 5.5 at a tenth of its predecessor's rates

Haiku 5.5 lists at $0.10 per million input tokens and scores 72.4% on OSWorld at maximum effort, but Anthropic estimates the real running-cost saving at about 75%, not 90%.

- Published: 2026-10-14T06:06:44.194Z
- Canonical: https://polylog.news/ai/2026-10-14/anthropic-prices-claude-haiku-5-5-at-a-tenth-of-its-predeces
- Publisher: Polylog (AI desk)
- Section: tech
- Sources: [Anthropic News](https://www.anthropic.com/claude-haiku-5-5), [Anthropic News](https://www.anthropic.com/claude-sonnet-5-5)

Anthropic released [Claude Haiku 5.5](https://www.anthropic.com/claude-haiku-5-5) on October 7 and called it its fastest, cheapest and most capable small model, built for high-volume, cost-sensitive work. The list price is [$0.10 per million input tokens and $0.50 per million output tokens](https://apidog.com/blog/claude-haiku-5-5-pricing/) for prompts up to 100,000 tokens. Above that threshold the price rises to $0.50 and $2.50. Cache reads start at $0.01 per million tokens, and the Batch API halves every rate.

The list price is one tenth of Haiku 4.5's. Anthropic's own estimate is that Haiku 5.5 costs [around 75% less to run](https://braindetox.kr/en/posts/claude_haiku_5_5_release_pricing_2026.html) on average. The gap between the two figures implies that the new model generates more tokens per task, especially at higher effort settings, so buyers should budget against the 75% figure.

On Anthropic's published results, [Haiku 5.5 scores 72.4% on OSWorld, 39.2% on Terminal-Bench and 1620 on GDPval-AA at maximum effort](https://computingforgeeks.com/claude-haiku-5-5-released-features-benchmarks/). The cost-per-attempt comparison matters most for agent builders. On OSWorld, Haiku 5.5 at maximum effort (72.4% at $0.61) beats [Claude Sonnet 5.5](https://www.anthropic.com/claude-sonnet-5-5) at low effort (57.9% at $0.68), [according to MetricNexus](https://metricnexus.ai/blog/claude-haiku-5-5-benchmarks-pricing). Haiku 5.5 at medium effort (53.3% at $0.13) beats OpenAI's GPT-6 Luna at maximum effort (48.9% at $0.20). Artificial Analysis gives Haiku 5.5 a score of [43 on its Intelligence Index](https://techjacksolutions.com/ai-brief/claude-haiku-5-5-benchmark-pricing-artificial-analysis/), slightly ahead of Gemini 3.8 Flash (41) and GPT-6 Luna (38).

The release follows Sonnet 5.5, launched September 28, which Anthropic says runs 30% faster and costs up to 30% less than Sonnet 5 for most work. Meta set the reference point in July, when it priced its [Muse Spark 1.1 API at $1.25 and $4.25](https://aiweekly.co/alerts/meta-prices-muse-spark-11-api-at-125425-per-m-tokens) per million tokens.

## What this means

Small-tier pricing now sets the cost of running agent loops, because high-volume steps such as browsing, extraction and tool calls run on these models. At $0.10 per million input tokens, Anthropic is pricing below most hosted open-weight endpoints. That reduces the main advantage of open-weight models on hosted inference platforms, and it pressures GPT-6 Luna and Gemini 3.8 Flash directly. Anthropic can absorb the cut only if inference efficiency gains have exceeded the price reduction, which its 75% running-cost estimate partly supports. The main beneficiaries are application companies with thin margins on token-heavy products. Inference resellers and GPU cloud providers lose pricing power on small-model traffic. Most coverage reads each cut as a single launch, but Sonnet 5.5, Haiku 5.5 and Google's Argon discount all came within about two weeks, which fits a sustained price war. The likelier outcome is that OpenAI and Google reprice their small tiers within weeks. That depends on Haiku 5.5 winning visible share in agent workloads. The read is wrong if rivals hold prices through the end of the year, which would mean they judge Haiku's capability lead too small to cost them customers.

## What to watch

- Whether OpenAI cuts GPT-6 Luna pricing in the coming weeks. A cut would confirm that list price is now the main competitive tool at the small tier.
- Independent token-count measurements for Haiku 5.5 at maximum effort. They will show whether the 75% running-cost saving holds on real agent workloads.
- Price changes from hosted open-weight providers. Cuts would show that Haiku 5.5 is drawing traffic away from open models.