Claude Haiku 5.5
Anthropic's third Claude 5.5 model — 75% cheaper to run than Haiku 4.5 at $0.10/$0.50 per 1M tokens under 100k context, with an adjustable effort setting (first for Haiku), 1M-token context, and built-in cybersecurity safeguards.
Why it matters
The fastest Anthropic tier resets the small-model price war: 72.4% on OSWorld 2.1 (vs 15.7% for Haiku 4.5) makes it a credible browser-use and subagent model at near-floor pricing.
What you could build with it
A startup building browser-based research agents could run Haiku 5.5 as the per-step subagent model for navigation and form-filling, reserving the frontier model for planning — cutting agent run costs by roughly three-quarters while keeping task completion rates.
Does it hold up?
Cautiously promising but evidence is thin: the only real usage data so far is vendor-supplied (six Anthropic-quoted enterprise pilots with latency/quality wins), and independent third-party production results are absent one day post-launch. The price (~75% cheaper than Haiku 4.5) makes it the obvious default for high-volume subagent/summarization work, but treat vendor benchmarks as directional until independent evals land.
Built with Claude Haiku 5.5
- Unite.AI: early enterprise testing of Haiku 5.5 (Asana, HubSpot, AlphaSense, Box, Cognition, Rogo)unite.ai · Six companies reported early production tests: Asana saw 30% lower latency and 2.5x faster per-agent-turn inference for its AI Teammates agent; HubSpot got its best-ever score (92.8%) on its simulated CRM suite; AlphaSense measured a statistically significant lift (0.84 vs 0.76) across 400 production-style queries at ~8M calls/week; Cognition ran Devin Fusion with Haiku 5.5 as sidekick model holding a 66.2 FrontierCode score.
- OpenRouter: live usage stats for claude-haiku-5-5openrouter.ai · Haiku 5.5 was available on OpenRouter within hours of launch (Oct 7, 2026) under model ID claude-haiku-5-5 with 1M context at $0.10/$0.50 per 1M tokens, plus a batch variant at $0.05/$0.25 — developers are routing traffic to it the same day it shipped.
- The Decoder: Haiku 5.5 arrives with massive price cutsthe-decoder.com · Benchmark recap plus an important caveat: the new tokenizer consumes ~30% more tokens per task, so real-world savings are smaller than the 90% per-token headline suggests.
Learn more
First spotted on article: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.