anthropic.com web signal

Anthropic Cuts Claude Haiku 5.5 Price 75% Below Haiku 4.5

By Published Updated
6 sources tracking this story
Anthropic Inference Agents ai-business

TL;DR

  • The 90% price cut applies only below 100k tokens; above that threshold, rates jump 5x to $0.50/$2.50 per million input/output tokens.
  • Artificial Analysis benchmarks clock Haiku 5.5 max mode at roughly 162k output tokens per task versus roughly 54k for GPT-6 Luna max, a 3x efficiency gap that compresses real-cost advantage on reasoning workloads.
  • A new tokenizer generates roughly 30% more tokens per character than Haiku 4.5, narrowing effective savings further than the 75% average headline implies.

Anthropic has priced its new small model at $0.10 per million input tokens and $0.50 per million output tokens for prompts up to 100,000 tokens. On the launch page, the company says Haiku 5.5 "is priced 90% lower than Claude Haiku 4.5 for requests up to 100,000 tokens, and 50% lower for requests over 100,000 tokens," and "on average, it now costs around 75% less to run."

The pitch is narrow. Anthropic calls Haiku 5.5 "the cheapest, fastest, and most capable small model we've ever released" and tells developers it is "designed for high-volume, cost-sensitive tasks" like summaries, compactions, database queries and classification. It is also, Anthropic writes, the first Haiku-class model with an adjustable effort setting, so "users can decide whether to optimize for cost or intelligence."

The reported scores: 72.4% on OSWorld 2.1 (offline subset), 45.9% on Humanity's Last Exam without tools and 57.4% with them, and 39.2% on Terminal-Bench 4.0. Anthropic positions the model as a subagent for Opus 5.5 and Sonnet 5.5 on coding work, which fits a run of agent-shaped stories moving through our agents tracker this quarter.

Two other changes shipped in the same post. Sonnet 5.5 cache reads drop to $0.10 per million tokens from $0.20, which Anthropic says "means Sonnet 5.5 now runs around 20% cheaper on most agentic work." And paid-plan users get monthly API credits: "Max 5x users will get $100 in credits per month, Max 20x users will get $200, and Team subscribers will receive up to $500, pooled across their users."

The one customer figure Anthropic puts on the page is from Asana. "We saw over a 30% reduction in latency for task completions and up to 2.5x faster inference per agent turn," says Aaron Vinh, a Staff Software Engineer there. The launch page carries no direct comparison to OpenAI or other named competitors.

What others are reporting

Coverage cluster as of 24h after publish

  1. VentureBeat Read →

    Adds customer validation from Box, Asana, and Rogo with specific throughput metrics; breaks down tiered pricing structure and flags the absence of tokens-per-second specs.

    It's accurate enough that we'd trust it there and fast and cheap enough that we can run it a lot.
  2. Artificial Analysis Read →

    Independent benchmark data quantifies the output-token efficiency gap versus GPT-6 Luna max and provides AA-Briefcase Elo, Terminal-Bench 4.0, and hallucination scores absent from Anthropic's release.

    Haiku 5.5 (max) uses ~162k output tokens per Intelligence Index task, ~3x GPT-6 Luna (max)
  3. Yahoo Finance Read →

    Frames the price cut as competitive necessity tied to Anthropic's pre-IPO positioning, citing $4.6B revenue against $42B in 2025 losses and a broader enterprise shift away from the tokenmaxxing era.

    Haiku 5.5 costs an average of 75% less than Haiku 4.5
  4. AWS Machine Learning Blog Read →

    First-party deployment context covering GovCloud availability, IAM and CloudTrail governance integration, and multi-region inference profiles across US, EU, AU, and JP.

    Claude Haiku 5.5 is the fastest and most efficient model in the Claude 5.5 family, built for subagents and high-volume, cost-sensitive work.
  5. DataCamp Read →

    Hands-on testing quantifies the tokenizer efficiency gap (30% more tokens than Haiku 4.5), explains the 90% vs 75% savings discrepancy, and documents the OSWorld 2.1 desktop-automation leap from 15.7% to 72.4%.

Shared on Bluesky by 3 AI experts

Originally reported by anthropic.com

Read the original article →

Original headline: Anthropic Ships Claude Haiku 5.5 at 75% Lower Cost, Pairs With Sonnet Cache Cut and Max API Credits