October 2026
State of AI Cost, October 2026
Published October 1, 2026. Covers Oct 1 to Oct 31, 2026.
The Frontier index sits at $4.64 per million tokens, unchanged. The lowest blended price on record is perplexity/pplx-decider-flash-v1.2 at $0.0075 per million tokens. 10 models changed price. The community tracked $6690.56 of AI spend across 5 contributors. 5 developers synced coding-agent usage worth $8286.42 at API list prices; the median developer's was $1.02.
Key takeaways
Frontier AI held flat this period, at $4.64 per 1M tokens (blended).
Biggest move: openrouter/deepseek/deepseek-v4.1-flash output price rose 300%.
Cheapest credible model: perplexity/pplx-decider-flash-v1.2 per 1M tokens (blended).
the median developer's coding-agent usage at API list prices, across 5 people.
The AI Cost Index
See the live AI Cost Index for the current value and full history.
Biggest price moves
| Model | Output before | Output after | Change |
|---|---|---|---|
| openrouter/deepseek/deepseek-v4.1-flash | $0.30 | $1.20 | +300% |
| tencent/deepseek-v4-pro | $0.87 | $3.48 | +300% |
| mistral/mistral-large-4 | $2.09 | $4.18 | +100% |
| mistral/mistral-large-4-0 | $2.09 | $4.18 | +100% |
| wandb/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B | $0.75 | $1.50 | +100% |
| azure/gpt-4o-mini-audio-preview-2024-12-17 | $10.00 | $0.60 | -94% |
| openrouter/~anthropic/claude-haiku-latest | $5.00 | $0.50 | -90% |
| fireworks_ai/accounts/fireworks/models/deepseek-v4p1-flash | $0.66 | $1.20 | +81.8% |
| fireworks_ai/deepseek-v4p1-flash | $0.66 | $1.20 | +81.8% |
| databricks/databricks-gpt-5-6-luna | $6.00 | $1.20 | -80% |
Lowest cost per million tokens
What the community spent
Top models by spend
Anonymized aggregate across opted-in contributors. See the live community dashboard.
What coding agents cost developers
5 developers synced their Claude Code and other coding-agent usage with npx mytokentracker this period. Priced at API list rates it came to $8,286.42. Anyone on a flat plan like Claude Pro or Max was billed the plan price instead, so read this as what the same usage would cost per token, not what anyone paid.
By agent
2 agents are not shown, so no one person's usage can be worked out from the totals.
By model
| Model | People | API-equivalent | Share |
|---|---|---|---|
| Claude Opus 5.5 | 5 | $133.42 | 2% |
13 models are not shown, so no one person's usage can be worked out from the totals.
A small sample, read it that way. These are 5 people who opted in to community numbers, with MyTokenTracker's own admin accounts left out. Rows shared by fewer than 3 people are dropped, not pooled, along with any row that would let a dropped one be worked out by subtraction. It describes what this group did, not an industry average. Methodology.
Cite this report
The numbers in this report are frozen at publication, so it is safe to cite as a fixed, dated snapshot.
Copy a citation
Free to use and cite under CC BY 4.0. See how this is measured.
Champlin Enterprises. (2026). State of AI Cost, October 2026 [Report]. MyTokenTracker. Retrieved November 13, 2026, from https://mytokentracker.io/reports/state-of-ai-cost-2026-10
@misc{mtt-report-state-of-ai-cost-2026-10,
title = {State of AI Cost, October 2026},
author = {{Champlin Enterprises}},
year = {2026},
howpublished = {MyTokenTracker, \url{https://mytokentracker.io/reports/state-of-ai-cost-2026-10}},
note = {Accessed November 13, 2026. Licensed CC BY 4.0.},
url = {https://mytokentracker.io/reports/state-of-ai-cost-2026-10}
}
Need a fixed point in time? Every day’s data is permanently archived in the open-data repository, so you can cite a specific date by linking that day’s committed file.