Claude Haiku 5.5 Pricing: What a 90% Cheaper Haiku Unlocks

Anthropic released Claude Haiku 5.5 on October 7, 2026, and the headline number is hard to ignore: for prompts up to 100,000 tokens, it costs $0.10 per million input tokens and $0.50 per million output tokens. That is a tenth of what Claude Haiku 4.5 cost, for a small model that Anthropic says now handles agentic computer-use and terminal tasks its predecessor could barely attempt. For anyone running high-volume AI workloads, this is the most consequential pricing change of the autumn.

• Pricing: $0.10 input / $0.50 output per million tokens up to 100K-token prompts; $0.50 / $2.50 above that. Haiku 4.5 was $1 / $5.
• Context and output: a 1M-token context window with up to 128K output tokens (300K on the Batch API with a beta header).
• Benchmarks (Anthropic): OSWorld 2.1 offline subset 72.4% vs 15.7% for Haiku 4.5; Terminal-Bench 4.0 39.2% vs 0.0%.
• Availability: Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry and Claude Platform on AWS, model ID claude-haiku-5-5.

What Is Claude Haiku 5.5 Built For?

Anthropic positions Haiku 5.5 for high-volume, latency-sensitive work: classification, routing, extraction, and, increasingly, acting as a subagent inside larger agent systems. It is the fastest model in the current Claude lineup and accepts text and images as input. Its reliable knowledge cutoff is June 2026, the same as the larger Claude 5.5 models.

The genuinely new trick for this tier is control. Haiku 5.5 is the first Haiku-class model with an adjustable effort setting, running from Low through Max, and adaptive thinking is on by default at medium effort. In practice, that means a developer can let the model answer a simple routing question almost instantly, then turn the dial up for a harder extraction job without switching models.

How Much Cheaper Is Haiku 5.5 Than Haiku 4.5?

Anthropic says the new model averages about 75% cheaper than Haiku 4.5: roughly 90% cheaper on prompts under 100,000 tokens and 50% cheaper above that line. Cache reads drop to $0.01 per million tokens on shorter prompts, and the Batch API halves input and output prices again.

One honest caveat from Anthropic's own documentation: Haiku 5.5 uses the newer tokenizer shared with recent Claude models, so the same text counts as about 30% more tokens than it did on Haiku 4.5. Real-world savings will therefore be somewhat smaller than the sticker comparison, though still substantial for most workloads. Teams migrating should re-measure their token counts rather than assume a straight 90% cut.

Haiku 5.5 Benchmarks and Early Customer Results

The capability jump is what makes the price story interesting. According to Anthropic's announcement, Haiku 5.5 scores 57.4% on Humanity's Last Exam with tools (Haiku 4.5: 18.7%) and an Elo of 1620 on GDPval-AA, a measure of real-world knowledge work, compared with 735 for Haiku 4.5 and 1437 for OpenAI's GPT-6 Luna. The larger Claude Sonnet 5.5 still leads on every benchmark listed, which is the expected shape of a model family.

Early customers quoted by Anthropic describe concrete gains:

• HubSpot reported a 92.8% average on its CRM evaluation suite, the best score it has seen there.
• Box said Haiku 5.5 scored 11 points higher than Haiku 4.5 at about half the latency.
• Asana reported up to 2.5x faster inference per agent turn.
• AlphaSense measured 0.84 versus 0.76 for Haiku 4.5 on 400 document-question queries.

Independent coverage has started to land. DataCamp's launch write-up and Help Net Security both confirm the pricing and benchmark figures, and a hands-on test from Classmethod confirms the model is live on Amazon Bedrock with regional inference profiles.

What Else Did Anthropic Change on Launch Day?

Alongside Haiku 5.5, Anthropic halved Claude Sonnet 5.5 cache-read pricing from $0.20 to $0.10 per million tokens, which it says lowers Sonnet costs on most agentic tasks by about 20%. Max and Team subscribers also get a new monthly API credit, $100 for Max 5x and $200 for Max 20x, and the Python and TypeScript SDKs add beta support for computer use and browser use.

Why This Matters for Builders

The pattern across this year has been big models getting smarter while small models get dramatically cheaper. Haiku 5.5 compresses both trends into one release: a model priced like a utility that can still operate a desktop or a terminal with real competence. If you have been weighing the bigger models, our Sonnet 5.5 vs Opus 5.5 comparison and Opus 5.5 pricing breakdown show where Haiku now fits in the lineup. For more model launches, browse our artificial intelligence coverage.

Sources: Anthropic — Claude Haiku 5.5 — October 7, 2026; Claude Platform docs — Haiku 5.5 overview — October 7, 2026; DataCamp — October 7, 2026; Help Net Security — October 8, 2026; Classmethod — October 8, 2026.