Who changed AI API prices in September 2026
· covers September 2026
18 host-model prices changed in September 2026: 5 up, 13 down. These are the providers’ own published figures, recorded daily.
Which AI API prices changed?
| Model | Host | Was | Now | Change |
|---|---|---|---|---|
| deepseek/deepseek-v4-flash-0731 | reka | $0.021 | $0.100 | +376.2% |
| qwen/qwen3.8-27b | reka | $0.040 | $0.150 | +275.0% |
| z-ai/glm-5.3-flash | reka | $0.060 | $0.120 | +100.0% |
| z-ai/glm-5.3-flash | deepinfra | $0.075 | $0.150 | +100.0% |
| reflection/beam-1.0-501b-a23b | baseten | $1.300 | $0.400 | -69.2% |
| deepseek/deepseek-v4.1-flash | novita | $0.195 | $0.120 | -38.5% |
| z-ai/glm-5.3-flash | siliconflow | $0.150 | $0.100 | -33.3% |
| deepseek/deepseek-v4.1-flash-1017 | fireworks | $0.300 | $0.210 | -30.0% |
| deepseek/deepseek-v4.1-flash | fireworks | $0.300 | $0.210 | -30.0% |
| z-ai/glm-5.3 | fireworks | $1.400 | $0.980 | -30.0% |
| z-ai/glm-5.3 | deepinfra | $0.562 | $0.450 | -20.0% |
| deepseek/deepseek-v4-pro-0813 | deepinfra | $1.300 | $1.040 | -20.0% |
| deepseek/deepseek-v4-flash-0731 | deepinfra | $0.060 | $0.048 | -20.0% |
| z-ai/glm-5.3-flash | inference-net | $0.120 | $0.108 | -10.0% |
| z-ai/glm-5.3 | novita | $0.700 | $0.630 | -10.0% |
| z-ai/glm-5.2 | siliconflow | $0.700 | $0.651 | -7.0% |
| moonshotai/kimi-k3 | inference-net | $0.249 | $0.234 | -6.0% |
| z-ai/glm-5.3 | inference-net | $0.080 | $0.084 | +5.0% |
Where do these numbers come from?
Prices are published by the providers and read from a public catalogue once a day; the change is the difference between the first and last reading in the month. We do not measure prices — we measure latency, which sits next to them on the price-vs-latency page.
Same series: Previous: August 2026 · Next: October 2026
Other measurements from the same period
- Which AI API SDKs developers installed in September 2026
- AI API outages in September 2026: measured vs. admitted
- AI models retired or deprecated in September 2026
This post is a snapshot. The live figures behind it: /price-vs-latency · all posts