Anthropic Just Dropped Claude Haiku 5.5 (Major Upgrade)

Summary
Brock Mesarich breaks down the release of Anthropic’s Claude Haiku 5.5, analyzing its official launch details, pricing structure, and benchmark performance. He explains the tiering of the Claude 5.5 model family, demonstrates how Haiku 5.5 functions as a fast and low-cost subagent, and reviews concurrent price cuts to Sonnet 5.5 caching alongside new monthly API credits.

What is shown

• Release Timeline & Lineup Overview [00:08]: Graphics displaying the Claude 5.5 family release dates (Opus 5.5 on Sep 22, Sonnet 5.5 on Sep 28, Haiku 5.5 on Oct 7) and hierarchy from Fable 5.1 down to Haiku 5.5.
• Anthropic Announcement & Pricing Breakdown [00:41 - 02:24]: Excerpts from Anthropic’s blog post and comparison graphics detailing token pricing, tiered prompt brackets (under/over 100k tokens), and cache pricing.
• Benchmark Comparisons [02:42 - 03:25]: Benchmark table comparing Haiku 5.5, Haiku 4.5, OpenAI’s GPT-6 Luna, and Sonnet 5.5 across OSWorld 2.1, GDPval-AA v2.1, and Terminal-Bench 4.0.
• Effort Setting UI [03:29]: An interface mockup showing an adjustable “Effort” slider (Faster vs. Smarter: Low, Medium, High, Extra high, Max) for controlling cost versus reasoning depth.
• Use-Case Framework & Early Feedback [03:45 - 05:46]: Infographics on model selection guidelines, subagent workflows (featuring Rogo’s 10-K extraction pipeline), and early enterprise test metrics from Asana and Box.
• Sonnet 5.5 Cache Cut & Subscription API Credits [05:51 - 06:29]: Anthropic documentation detailing the 50% price reduction on Sonnet 5.5 cache reads and monthly developer API credits for Claude Max and Team tiers.

Claims & numbers

• Pricing & Token Limits:
  • Haiku 5.5 costs on average ~75% less to run than Haiku 4.5.
  • For prompts up to 100k tokens (about 55,000 words, representing ~90% of requests), input pricing dropped from $1.00 to $0.10 per million tokens and output pricing from $5.00 to $0.50 per million tokens (90% reduction).
  • For prompts exceeding 100k tokens, prices are $0.50 input and $2.50 output per million tokens (50% cheaper than Haiku 4.5).
  • Compared to Sonnet 5.5 ($2 input / $10 output), base Haiku 5.5 is 20 times cheaper.
• Benchmarks:
  • OSWorld 2.1 (Computer use): Haiku 5.5 scored 72.4%, compared to Haiku 4.5 (15.7%), GPT-6 Luna (48.9%), and Sonnet 5.5 (83.9%).
  • GDPval-AA v2.1 (Knowledge work): Haiku 5.5 scored 1620, more than double Haiku 4.5 (735), outperforming GPT-6 Luna (1437), and trailing Sonnet 5.5 (1840).
  • Terminal-Bench 4.0 (Agentic coding): Haiku 5.5 scored 39.2%, compared to Haiku 4.5 (0.0%), GPT-6 Luna (16.4%), and Sonnet 5.5 (70.6%).
• Early Tester Claims:
  • Asana reported a >30% reduction in latency for task completions and up to 2.5x faster inference.
  • Box reported Haiku 5.5 scored 11 points higher than Haiku 4.5 at approximately half the latency.
• Additional Updates:
  • Claude Sonnet 5.5 cache read pricing was halved from $0.20 to $0.10 per million tokens, reducing Sonnet 5.5 agentic task costs by ~20%.
  • Monthly API credits introduced for subscribers: $100/month for Max 5x, $200/month for Max 20x, and up to $500/month pooled for Team plans.

Notable quotes

• “Anthropic just dropped their cheapest model in the Claude 5.5 family, and it could change which model you use for a lot of your everyday work.” [00:00]
• “So if we do the math here, that is 90% lower.” [01:46]
• “Anthropic says Haiku pairs really well with Opus and Sonnet as a subagent on coding work.” [05:02]

Assessment
This is a creator commentary and analytical breakdown summarizing Anthropic’s official blog announcements, pricing sheets, and evaluation tables. The video does not include live runtime coding or prompt executions, relying entirely on presentation slides, graphics, and official quotes.

Described by gemini-3.8-flash on 2026-10-10 from the video’s audio and frames.