DeepSeek vs Llama
The short answer. The cheapest DeepSeek model we track, deepseek-flash, costs $1.20 per million output tokens and $0.300 per million input tokens. The cheapest Llama model, Llama 3.1 8B Instruct, costs $0.080 per million output tokens and $0.050 per million input tokens, as sold by OpenRouter. At the entry level, Llama is 15.0 times cheaper than DeepSeek on output tokens. Prices read on 11 September 2026 and 18 September 2026.
This page compares what DeepSeek and Meta charge through their APIs. Every price below carries the day we read it and links to the page it came from. We do not rank quality: we tell you what each model costs, and since when.
At the top of each range
- Among DeepSeek models that carry a relayed score, the highest on the Artificial Analysis intelligence index is DeepSeek V4.1 Flash, at 39.5, sold by OpenRouter at $0.600 per million output tokens.
- Among Llama models that carry a relayed score, the highest on the Artificial Analysis intelligence index is Llama 4 Maverick, at 9.3, sold by OpenRouter at $0.6525 per million output tokens.
The index is produced by Artificial Analysis and relayed through OpenRouter's catalogue. It is not our measurement; we record it as a level 4 source and show it next to the price we read. A model without a score here has not been scored yet, which is not the same as scoring low.
Every model, side by side
DeepSeek models
2 standard offers, from the prices DeepSeek publishes itself.
| Model | Input | Output | Context | Read on | Source |
|---|---|---|---|---|---|
| deepseek-flash | $0.300 | $1.20 | 1 000 000 | 11 September 2026 | DeepSeek |
| deepseek-v4-pro | $1.32 | $3.96 | 1 000 000 | 11 September 2026 | DeepSeek |
Llama models
8 standard offers, from OpenRouter's prices: we track no price list published by Meta itself for these models.
| Model | Input | Output | Context | Read on | Source |
|---|---|---|---|---|---|
| Llama 3.1 8B Instruct | $0.050 | $0.080 | 131 072 | 18 September 2026 | OpenRouter |
| Llama Guard 4 12B | $0.180 | $0.180 | 163 840 | 18 September 2026 | OpenRouter |
| Llama 3.2 1B Instruct | $0.027 | $0.201 | 60 000 | 18 September 2026 | OpenRouter |
| Llama 4 Scout | $0.100 | $0.300 | 1 310 720 | 18 September 2026 | OpenRouter |
| Llama 3.3 70B Instruct | $0.100 | $0.320 | 131 072 | 18 September 2026 | OpenRouter |
| Llama 3.2 3B Instruct | $0.050 | $0.330 | 131 072 | 18 September 2026 | OpenRouter |
| Llama 3.1 70B Instruct | $0.400 | $0.400 | 131 072 | 18 September 2026 | OpenRouter |
| Llama 4 Maverick | $0.1875 | $0.6525 | 1 048 576 | 18 September 2026 | OpenRouter |
How these prices have moved
We hold dated captures of these models' prices back to 19 June 2024. We did not observe these values ourselves: they were recovered from public captures of OpenRouter's catalogue, and each links to its capture so you can check it without trusting us. Largest moves on output price:
| Model | Before | After | Change | Captured |
|---|---|---|---|---|
| Meta: Llama 3.2 1B Instruct | $0.010 | $0.200 | +1900% | 14 November 2025 |
| Meta: Llama 3.2 3B Instruct | $0.020 | $0.340 | +1600% | 3 March 2026 |
| DeepSeek: R1 Distill Llama 70B | $0.110 | $0.800 | +627% | 19 February 2026 |
| DeepSeek: DeepSeek V4 Pro 0813 | $0.870 | $3.564 | +310% | 21 August 2026 |
| Meta: Llama 3.2 3B Instruct | $0.006 | $0.024 | +300% | 4 September 2025 |
| Meta: Llama Guard 4 12B | $0.050 | $0.180 | +260% | 12 August 2025 |
Questions
Which is cheaper, DeepSeek or Llama?
The cheapest DeepSeek model we track, deepseek-flash, costs $1.20 per million output tokens and $0.300 per million input tokens. The cheapest Llama model, Llama 3.1 8B Instruct, costs $0.080 per million output tokens and $0.050 per million input tokens, as sold by OpenRouter. At the entry level, Llama is 15.0 times cheaper than DeepSeek on output tokens. Prices read on 11 September 2026 and 18 September 2026.
What is the cheapest DeepSeek model?
deepseek-flash, at $1.20 per million output tokens, read on 11 September 2026.
What is the cheapest Llama model?
Llama 3.1 8B Instruct, at $0.080 per million output tokens, read on 18 September 2026.
Where do these prices come from?
From the price lists DeepSeek and OpenRouter publish, read by our collectors. No figure on this page was produced by a language model.