Anthropic launches Claude Haiku 5.5: 75% cheaper, 39.2% on Terminal-Bench vs GPT-6 Luna's 16.4%
dl_weekly · x · 2026-10-17
Anthropic released Claude Haiku 5.5, its cheapest, fastest and most capable small model, priced at $0.10/$0.50 per million tokens below 100K context — roughly 75% cheaper to run than Haiku 4.5.
- Scores 39.2% on Terminal-Bench 4.0 at max effort, far above GPT-6 Luna's 16.4%; 45.9% on Humanity's Last Exam without tools (57.4% with tools); 72.4% on OSWorld 2.1 offline subset
- Positioned for high-volume workloads (summaries, compaction, classification), as a coding subagent alongside Opus 5.5 / Sonnet 5.5, and for latency-sensitive uses like live support and browser use
- Sonnet 5.5 cache reads halved in price (20% cheaper on most agentic work); Claude Max and Team subscribers get a new monthly API credit
More from coding & agent
- fx v0.0.14 ships durable agents: same API survives crashes on laptop, cloud via Vercel — cramforce · 2026-10-17
- Jeffrey Emanuel ships agent skill turning audio into spectrogram karaoke videos — doodlestein · 2026-10-17
- OpenCall playbook: agent swarm does on-call triage with human approval and auto fix PRs — please-dont-deploy · 2026-10-17
- Spotify MCP Server lets AI assistants like Cursor and Claude control playback — modelcontextprotocol · 2026-10-17
- pqs-mcp-server scores any prompt across 8 dimensions before the model runs — modelcontextprotocol · 2026-10-17
- AI Industry Worker Says Nobody He Knows Uses xAI's Grok Build — SimonBalmain · 2026-10-17