Anthropic Releases Haiku 5.5 with 75% Cost Cuts and Performance Gains

Anthropic unveiled Claude Haiku 5.5 on October 7, 2026, a compact language model offering 75% lower costs than its predecessor Haiku 4.5 while significantly improving performance on reasoning and coding tasks. The model is now available across major platforms and introduces an adjustable effort setting for cost or accuracy optimization.

Quick Facts

Released Claude Haiku 5.5 language model
Reduced pricing by 75% average on operational costs
Improved performance on benchmarks including OSWorld 2.1, Terminal-Bench 4.0, and knowledge work tasks
Added adjustable effort setting for Haiku-class models
Halved Claude Sonnet 5.5 cache read pricing

Anthropic released Claude Haiku 5.5 on October 7, 2026, as a compact language model designed for high-volume, cost-sensitive tasks. The model delivers approximately 75% lower operational costs compared to Haiku 4.5, with pricing reduced to $0.10 per million input tokens and $0.50 per million output tokens for requests with prompts up to 100,000 tokens—which account for roughly 90% of Haiku 4.5 requests. For requests exceeding 100,000 tokens, pricing is $0.50 input and $2.50 output per million tokens, representing 50% savings versus the previous generation.

Haiku 5.5 demonstrates significant performance improvements on benchmark tasks. On OSWorld 2.1 (a test measuring computer operation across multiple steps), it achieves 72.4% accuracy compared to Haiku 4.5's 15.7% and outperforms GPT-6 Luna's 48.9%. On Terminal-Bench 4.0 (measuring command-line coding tasks), it scores 39.2% versus Haiku 4.5's 0.0%, though it remains behind Sonnet 5.5's 70.6%. Knowledge work benchmarks (GDPval-AA v2.1 and AA-Briefcase v1.1) show Haiku 5.5 at 1620 and 1578 Elo points respectively, compared to Haiku 4.5's 735 and 614.

The model is positioned as a subagent for coding work alongside Opus 5.5 and Sonnet 5.5, and is noted as the fastest model in Anthropic's lineup (excluding Opus's Fast Mode). It is designed for repetitive tasks including summaries, context compression, database queries, and classification, as well as speed-critical applications like live customer support and browser automation.

Haiku 5.5 is the first Haiku-class model to support an adjustable effort setting, allowing users to balance cost and reasoning depth across five levels. The model's knowledge cutoff is June 2026. Alignment evaluations show major improvements across nearly all metrics relative to Haiku 4.5, with fewer instances of misaligned behavior. Cybersecurity safeguards are more restrictive than Haiku 4.5 but permit a wider range of defensive tasks than applied to Sonnet 5.5.

The model is immediately available across all platforms, including the Claude app (free and paid tiers), Claude API, Amazon Web Services, Google Cloud, and Microsoft Azure. Developer access uses the model identifier "claude-haiku-5-5". Alongside the launch, Anthropic halved Claude Sonnet 5.5's cache read pricing from $0.20 to $0.10 per million tokens, reducing costs on agentic work by approximately 20%. The company also introduced monthly API credits for Claude Max and Team subscribers: $100 for Max 5x, $200 for Max 20x, and up to $500 monthly for Team accounts.

Topics

Technology Tech Breakthrough Business Artificial Intelligence

#Claude Haiku 5.5 #cloud computing #artificial intelligence #performance benchmark #pricing #Anthropic #AI #language model #Haiku 5.5 #AI pricing

Entities

Why This Matters

Developers and organizations using language models face direct cost and capability trade-offs. A 75% price reduction on a compact model while improving reasoning and coding benchmark performance may shift adoption patterns for cost-sensitive applications like customer support automation, database queries, and browser automation. Pricing changes to cache read operations on Sonnet 5.5 (down 50%) affect agentic AI workflows. Availability across AWS, Google Cloud, Azure, and native Claude API creates platform flexibility. Users building high-volume systems can now measure cost per task against performance gains on specific benchmarks (OSWorld 2.1, Terminal-Bench 4.0) to assess ROI.

Timeline & Sources

Sep 1, 2026

Wire

Anthropic released Claude Opus 5.5 and Claude Sonnet 5.5

Oct 7, 2026

Wire

Anthropic releases Claude Haiku 5.5

Oct 7, 2026

Wire

Sonnet 5.5 cache-read pricing reduced from $0.20 to $0.10 per million tokens

Oct 8, 2026

Wire

Media coverage and announcements in Japanese and Chinese publications

Quick Facts

Topics

Entities

Why This Matters

Timeline & Sources

Related Signals

Sources