October 2026 was full of AI releases, but the real headline was price. Anthropic's Claude Haiku 5.5 launched at US$0.10 per million input tokens — about 75% cheaper to run than its predecessor — and cached prices for Sonnet 5.5 halved. Google opened Gemini 4 Argon to developers at US$2 and US$10 per million tokens. Overall, running AI costs roughly ten times less than a year ago.
The price of the model was never the main cost of automation — describing the process is. But price decides which jobs are worth doing at all. A small task that runs a few hundred times a month may now pay back in months.
The smart approach is tiers: cheap, fast models for sorting and extracting; mid-size models for drafting; the most powerful models only for hard judgement calls.
What it means for your business
- Revisit the list of tasks that were 'too small to automate'.
- Use different model sizes for different tasks instead of one for everything.
- Measure cost per completed task — the number the business understands.
How MTH can help: Our AI Adoption service chooses the right model from OpenAI, Anthropic, Google or xAI for each use case, with a clear cost model.
This briefing is original commentary by MTH based on public reporting; figures are as reported by the sources named above and may be revised over time.
