State of AI pricing
State of AI pricing: August 2026
The ByteCosts State of AI Pricing for August 2026 tracks 193 AI providers and 6,821 priced models in one source-backed index. First-party list prices this period run from $0.140 to $3.00 per 1M input tokens, with a median of $0.600; output tokens cost $0.300 to $15.00 per 1M. The change ledger recorded 16 increases, 31 decreases, 337 new listings, and 40 removals. 57 models are offered by two or more providers; 8 cost at least double on the priciest host, and the widest gap is Kimi K2.6 at 956%. Not enough benchmarked rows yet to report value-per-dollar. These are list prices; verify the provider source before a billing decision.
Open the pricing index - Browse every model, source, and confidence grade →
The market at a glance
Snapshot generated August 22, 2026, from the committed ByteCosts dataset:
- 193 providers and 7,243 models tracked, 6,821 with a listed price.
- 0 first-party models carry an Artificial Analysis intelligence score.
- Median first-party input price $0.600 per 1M tokens; median output $2.50.
- Blended (70/30) list price runs $0.182 at the 10th percentile to $6.60 at the 90th.
Price distribution (first-party list prices)
Percentiles of input, output, and 70/30 blended token price across first-party lab models with a positive listed rate.
| Percentile | Input / 1M | Output / 1M | Blended / 1M |
|---|---|---|---|
| 10th | $0.140 | $0.300 | $0.182 |
| 50th (median) | $0.600 | $2.50 | $1.14 |
| 90th | $3.00 | $15.00 | $6.60 |
What moved this period
16 increases, 31 decreases, 337 new listings, and 40 removals in the committed change ledger. Biggest moves first.
| Model | Field | Change |
|---|---|---|
| Largest increases | ||
| DeepSeek deepseek-v4-pro | input | +11900% |
| DeepSeek deepseek-v4-flash | input | +4900% |
| DeepSeek deepseek-v4-pro | cacheRead | +1114% |
| DeepSeek deepseek-v4-pro | cachedInput | +1114% |
| DeepSeek deepseek-v4-flash | cacheRead | +400% |
| Largest decreases | ||
| Zhipu AI GLM-5.1 | cacheRead | -80% |
| DeepSeek deepseek-v4-pro | cachedInput | -75% |
| OpenCode Go MiniMax M3 | cacheRead | -67% |
| MiniMax (minimax.io) MiniMax-M3 | cacheRead | -67% |
| MiniMax (minimax.io) MiniMax-M3 (Priority, >512k input) | cacheRead | -67% |
The two findings behind the headline
The full rankings are on their own pages; this is the summary:
- 57 models are offered by two or more providers; 8 cost at least double on the priciest host, and the widest gap is Kimi K2.6 at 956%. See /model-price-spread for the full table.
- Not enough benchmarked rows yet to report value-per-dollar.
How this report is built
Every number is derived at build time from the committed ByteCosts dataset: no live API, no hand-edited figure. The market headline uses the dataset summary counts; the price distribution uses first-party lab models with a positive listed rate; the change summary uses the append-only price-event ledger.
A price only appears in the ledger after the data is refreshed manually and committed, so the report is never ahead of the data. Where the log contains only new listings, the report says so instead of inventing a trend.
Frequently asked questions
How is this different from the pricing index?
The pricing index is the browsable, filterable catalog. This report is the periodic summary: the market headline, the price distribution, what moved, and the standout findings, in one citable page that updates on each data refresh.
How often is the report updated?
Each data refresh re-stamps the snapshot (currently August 22, 2026). There is no live API; the numbers are list prices and can change between refreshes, so cite the snapshot date with any figure.
ByteCosts State of AI Pricing: August 2026. ByteCosts. Updated August 22, 2026. https://bytecosts.com/report/