Pangram charges ~$350/M input tokens; Anthropic Haiku does same job at 0.03% of the price
AaronBergman18 · x · 2026-11-08
Aaron Bergman breaks down Pangram's pricing: $0.05 per 100 words ≈ $350/M input tokens when processed in chunks. By comparison, Claude Haiku 5.5 processes the same short inputs at roughly $0.10/M tokens — 0.03% of the cost — while giving you a whole general model instead of a classifier.
He argues that unless there's a deep reason why tokenized text → probability should be orders of magnitude more expensive than LLM inference, an almost-as-good competitor could be built way below Pangram's price, and invites interested collaborators.
His broader point: OOM pricing drops unlock new categories — just as cheap LLMs enabled coding agents and swarms running for hours instead of 1:1 chats.
More from Venture
- Thinking Machines Lab offers $400K-$500K for SF infra research engineer with visa sponsorship — windx0303 · 2026-11-08
- Growth channel tier list: influencer marketing strong, cold email and LinkedIn ads bottom — LuoSays · 2026-11-08
- 2026 GEO Market Map: 145 Companies Across 10 Layers, Profound Leads at $1.8B — FinanceYF5 · 2026-11-08
- Small VC funds chase megafund-friendly deals for paper markups, says investor — arian_ghashghai · 2026-11-08
- Investor warns AI lab IPOs could bankrupt employees without tax-aware lockup planning — HanchungLee · 2026-11-08
- Is forward PE useless for valuing AI stocks? Two technologists debate the bottom — tengyanAI · 2026-11-08