Systems on Glass is an AI-automation side project of mine. Every episode is generated by Claude (Anthropic's AI, working through Claude Code in my own video studio): the script, the animations and 3D scenes, the edit, and the narration in an AI clone of my voice. I choose the topics, bring the stories from my own work, and review each episode before it goes up. Not affiliated with Anthropic or OpenAI.
The voice in my last episode cost fifty-two cents. I know because my studio prints the bill before it spends anything: 29 lines, 5,219 characters, $0.52. Most people using AI only see the bill at the end of the month. So this episode prints one.
The short version
- Two meters. Every call is billed for tokens in (instructions, files, the whole chat so far) and tokens out (the answer), at different rates. A rough rule: 1,000 tokens is about 750 English words, and every model counts its own way.
- Output costs about five times input. On Anthropic's list, Claude Opus 5.5 is $4 per million tokens in and $20 out; OpenAI's flagship is $10 in and $50 out (November 2026). The prompt is read in one pass; the answer is written one token at a time.
- In one call, the answer is the bill. 2,000 tokens in and 800 out comes to 2.4 cents: the answer is 29% of the tokens and 67% of the bill.
- In an agent, re-reading is the bill. Every turn re-sends the whole conversation. Thirty turns of a modest agent sends 1.17 million tokens in and only 15,000 out: $4.98, with re-reading at 94%.
- Pay less. Cache what repeats (a cache hit on Opus 5.5 is $0.20 per million, a twentieth: the same session drops to $0.86), batch what can wait (50% off at both labs), send easy work to a small model (Haiku 5.5 writes for a fortieth of Opus), ask for shorter answers, and log tokens per feature.
Agent, 30 turns (10,000 tokens to start, +2,000 a turn, 500 out a turn): 1,170,000 in = $4.68 · 15,000 out = $0.30 · total $4.98
Cached: 1,102,000 hits × $0.20/M + 68,000 writes × $5/M + $0.30 = $0.86 (−83%)
The call and the agent session are illustrative arithmetic on Anthropic's November 2026 list prices for Claude Opus 5.5; prices change, so check the live pages. The $0.52 is my studio's real estimate for episode 6's narration.
Further reading
- Anthropic: Pricing — per-model input, output, cache and batch prices.
- Anthropic: Prompt caching — how a repeated prefix is kept warm, and what writes and hits cost.
- OpenAI: API pricing — the same two-meter shape on another lab's list.
- OpenAI: Batch API — 50% off for jobs that can wait up to 24 hours.
- Context Is RAM — episode 4: why a smaller context is also a better one.
Next episode: Queues, and why my 7 AM starts at 6:00. Previously: Prompt Like a Spec. All episodes →