The biggest recent price moves are split hard between a few models getting much more expensive and a few getting dramatically cheaper. On the upside, openrouter/~deepseek/deepseek-pro-latest rose 230% from $0.6 to $1.98/Mtok blended, while on the downside openrouter/deepseek/deepseek-v4.1-flash fell 82.1% from $0.6 to $0.11/Mtok blended.
For developers, this is a reminder that model routing costs can change fast, especially on OpenRouter listings. If you are using any of these models in production, your per-request budget can swing meaningfully without any code change.
The largest increase in the set is openrouter/~deepseek/deepseek-pro-latest at +230%, followed by openrouter/~deepseek/deepseek-flash-latest at +150% and openrouter/moonshotai/kimi-k2.6 at +118.8%. That is not a small repricing, it is a reset of the cost baseline for those models.
There is also a clear mid-tier jump cluster, with openrouter/~z-ai/glm-flash-latest up 100% and openrouter/meta-llama/llama-3.3-70b-instruct up 87.1%. For teams that chose these models partly on price, the economics are now materially different, especially if they are used at scale.
On the cheaper side, openrouter/deepseek/deepseek-v4.1-flash dropped 82.1%, openrouter/openai/gpt-oss-120b:batch dropped 78.6%, and openrouter/deepseek/deepseek-v4-pro dropped 73.3%. That kind of decline can make a model newly viable for workloads that were previously too expensive, or let existing users expand usage without increasing spend.
The pattern here is simple: blended pricing is volatile, and the biggest budget risk is not just model choice, it is model churn. If you are tracking token spend closely, these moves are exactly the kind that can turn a stable monthly estimate into a surprise.
| Model | Provider | Was | Now | Change |
|---|---|---|---|---|
| openrouter/~deepseek/deepseek-pro-latest | Openrouter | $0.6 | $1.98 | +230% |
| openrouter/~deepseek/deepseek-flash-latest | Openrouter | $0.21 | $0.52 | +150% |
| openrouter/moonshotai/kimi-k2.6 | Openrouter | $0.78 | $1.71 | +118.8% |
| openrouter/~z-ai/glm-flash-latest | Openrouter | $0.12 | $0.24 | +100% |
| openrouter/meta-llama/llama-3.3-70b-instruct | Openrouter | $0.16 | $0.29 | +87.1% |
| openrouter/deepseek/deepseek-v4.1-flash | Openrouter | $0.6 | $0.11 | -82.1% |
| openrouter/openai/gpt-oss-120b:batch | Openrouter | $0.26 | $0.06 | -78.6% |
| openrouter/deepseek/deepseek-v4-pro | Openrouter | $0.98 | $0.26 | -73.3% |
- Watch openrouter/~deepseek/deepseek-pro-latest, it moved from $0.6 to $1.98/Mtok blended.
- openrouter/deepseek/deepseek-v4.1-flash and openrouter/openai/gpt-oss-120b:batch got much cheaper, which can reopen them for cost-sensitive workloads.
- openrouter/moonshotai/kimi-k2.6, openrouter/~z-ai/glm-flash-latest, and openrouter/meta-llama/llama-3.3-70b-instruct all saw large upward repricing.
- If your app depends on one model family, treat pricing as a moving input, not a fixed constant.
About this data. Figures are from MyTokenTracker live data (model prices, AI Cost Index, open data API), free to use and cite under CC BY 4.0. Snapshot: October 8, 2026.