How to read the table
Each row is one model tier from one provider. “Cached input” is the price of input tokens the provider has already seen in a recent request, such as a long system prompt. “Batch discount” is what you save by sending work that can wait up to a day.
I group the models into flagship, mid-size and small tiers because that is how most teams choose: pick a tier for the job, then compare providers inside it.
Prices per million tokens
Fourteen models in three tiers. The cheapest price in each column is easy to spot; the change column says what moved since October.
| Provider | Tier | Context | Input | Cached input | Output | Batch discount | Since October |
|---|---|---|---|---|---|---|---|
| Flagship models | |||||||
| Anthropic | Flagship | 200k | $5.00 | $0.50 | $25.00 | 50% | No change |
| OpenAI | Flagship | 400k | $4.00 | $0.40 | $16.00 | 50% | Output down from 20.00 |
| Flagship | 1M | $2.50 | $0.25 | $15.00 | 50% | No change | |
| Mistral | Flagship | 128k | $2.00 | $0.20 | $6.00 | None | No change |
| DeepSeek | Flagship | 128k | $0.60 | $0.060 | $2.20 | Off-peak 50% | Input up from 0.55 |
| Mid-size models | |||||||
| Anthropic | Mid | 200k | $3.00 | $0.30 | $15.00 | 50% | No change |
| OpenAI | Mid | 400k | $1.25 | $0.12 | $10.00 | 50% | No change |
| Mid | 1M | $0.30 | $0.030 | $2.50 | 50% | No change | |
| Mistral | Mid | 128k | $0.40 | $0.040 | $2.00 | None | New this month |
| Small models | |||||||
| Anthropic | Small | 200k | $1.00 | $0.10 | $5.00 | 50% | No change |
| OpenAI | Small | 400k | $0.25 | $0.025 | $2.00 | 50% | Input down from 0.30 |
| Small | 1M | $0.10 | $0.010 | $0.40 | 50% | No change | |
| Mistral | Small | 128k | $0.10 | $0.010 | $0.30 | None | No change |
| DeepSeek | Small | 128k | $0.27 | $0.070 | $1.10 | Off-peak 50% | No change |
Sample values for this layout. Context is the largest input the model accepts, in tokens.
What a month of work costs
A team that sends 10 million input tokens and 2 million output tokens a month, with the same long system prompt on most requests, so about 70% of input is cached.
| Provider | Tier | List price | With 70% cached | Saved |
|---|---|---|---|---|
| Flagship models | ||||
| Anthropic | Flagship | $100.00 | $68.50 | 31% |
| OpenAI | Flagship | $72.00 | $46.80 | 35% |
| Flagship | $55.00 | $39.25 | 29% | |
| Mistral | Flagship | $32.00 | $19.40 | 39% |
| DeepSeek | Flagship | $10.40 | $6.62 | 36% |
| Mid-size models | ||||
| Anthropic | Mid | $60.00 | $41.10 | 32% |
| OpenAI | Mid | $32.50 | $24.62 | 24% |
| Mid | $8.00 | $6.11 | 24% | |
| Mistral | Mid | $8.00 | $5.48 | 31% |
| Small models | ||||
| Anthropic | Small | $20.00 | $13.70 | 32% |
| OpenAI | Small | $6.50 | $4.92 | 24% |
| Small | $1.80 | $1.17 | 35% | |
| Mistral | Small | $1.60 | $0.97 | 39% |
| DeepSeek | Small | $4.90 | $3.50 | 29% |
Computed from the sample prices above: list price is 10 × input + 2 × output; the cached column bills 70% of input at the cached rate.
What changed since October
Three prices moved and one model is new. None of the changes reorders a tier, so if your provider was the cheapest in its tier in October, it still is.
Method
I read each provider’s public pricing page on the first working day of the month and copy the standard, pay-as-you-go rates. Volume deals, regional prices and free tiers are left out.