Recommended Articles
Tech & AI

Anthropic Launches Claude Haiku 5.5: Its Cheapest, Fastest Small Model Yet

By Angeled · October 8, 2026 · Tech & AI
Anthropic Launches Claude Haiku 5.5: Its Cheapest, Fastest Small Model Yet
Anthropic Launches Claude Haiku 5.5: Its Cheapest, Fastest Small Model Yet

What happened

Anthropic released Claude Haiku 5.5 on October 7, 2026 — the third model in the Claude 5.5 family to ship in roughly a month, following Opus 5.5 and Sonnet 5.5. The company describes it as "the cheapest, fastest, and most capable small model" it has ever released, designed for high-volume, cost-sensitive workloads.

Announcing the launch on X, Anthropic said: "Introducing Claude Haiku 5.5: the cheapest, fastest, and most capable small model we've ever released. On average, it costs around 75% less to run than Claude Haiku 4.5."

The model is available immediately on the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure under the model ID claude-haiku-5-5.

Key details

Pricing and the 75% cost claim. Anthropic says average running costs are around 75% below Haiku 4.5, factoring in tiered pricing and an updated tokenizer. Widely reported figures put pricing at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens (with cache reads at $0.01 and cache writes at $0.125), with rates quintupling above the 100K-token threshold — although some trackers note these per-token figures surfaced in pre-launch discussions and should be treated as unconfirmed until Anthropic publishes official pricing. Roughly 90% of requests to Haiku 4.5 fell within the cheaper tier, so the discount lands where most real workloads live. The launch also reduced Sonnet 5.5 cache-read pricing, making Sonnet 5.5 roughly 20% cheaper for most agentic workloads.

What it's built for. High-volume, repetitive, cost-sensitive work: summaries, context compaction, classification, database queries, customer support, voice agents, browser use, and in-app assistants. It can also serve as a subagent alongside Sonnet 5.5 or Opus 5.5 — the larger model handles complex reasoning while Haiku takes care of quick lookups and repetitive background tasks.

Adjustable effort setting. A first for the Haiku line: developers can tune an effort parameter to trade cost against intelligence, dialing the model up for harder sub-tasks or down for pure throughput.

Benchmarks. Anthropic reports 72.4% on the offline subset of OSWorld 2.1 (up from 15.7% for Haiku 4.5), 39.2% on Terminal-Bench 4.0 (where its predecessor scored zero), 46.4% on FrontierCode 1.1, and 45.9% on Humanity's Last Exam without tools (57.4% with tools). Digit.in reports Anthropic's comparisons showing Haiku 5.5 outperforming GPT-6 Luna on several of these benchmarks, including 46.4% vs 42.4% on FrontierCode 1.1 and 72.4% vs 48.9% on OSWorld. The model outputs text only, and its knowledge cutoff is June 2026.

Safety. Haiku 5.5 ships with built-in safeguards for a narrow set of high-risk cybersecurity requests — Anthropic's first Haiku model with such protections. Most everyday tasks are unaffected, but the move signals that even the cheapest tier is being hardened as AI agents take on more autonomous work.

Why it matters for founders and investors

The API pricing war that has run across every major AI lab since late 2025 just escalated again, and the beneficiary is every startup with an API bill. For founders building on model APIs, the math is straightforward: roughly 90% of Haiku-class requests fall under the 100K-token threshold, so a well-architected Haiku 5.5 deployment can cut inference costs dramatically versus the previous generation — money that flows directly into gross margin for AI-native products.

The more strategic shift is the multi-model architecture becoming the default. The intended pattern — Opus or Sonnet for hard reasoning, Haiku 5.5 for volume work — is now a first-class Anthropic design: a subagent tier with an adjustable cost/intelligence dial. Founders who architect their agents as tiered model pipelines from day one will out-compete those hard-coding a single model, because the cost floor keeps falling while capability rises. If your startup's moat is "we run expensive models," that moat just shrank.

For investors, Haiku 5.5's computer-use jump (15.7% to 72.4% on OSWorld) is the number to watch: cheap models are getting capable enough to drive the browser-use and agentic-workflow startups that dominate this year's AI funding rounds. Unit economics for AI agents improve every time a small model gets faster and cheaper — and this launch moves that curve materially.

Angeled.com connects founders with investors.

Source: TestingCatalog
Source: Unite.AI
Source: YourStory

Read Next