DeepInfra vs Groq.
Prices checked
DeepInfra is cheaper on 7 of the 7 models both sell, the widest gap Llama 3.3 70B Instruct at $0.16 against $0.64 per M tokens, checked .
Where to start
DeepInfraall 126 models Serves open-weight models on its own GPUs, often at fp8 or fp4, and publishes the precision per model.
Cheapest of the shared models
2.5¢/M tokens
Llama 3.1 8B Instruct
Groqall 9 models Serves open-weight models on its own LPU hardware with a free tier.
Cheapest of the shared models
5.8¢/M tokens
Llama 3.1 8B Instruct
Side by side
| Fact | DeepInfra | Groq |
|---|---|---|
| Kind of provider | Host | Host |
| Models we track | 126 | 9 |
| Cheaper of the two | 7 of 7 | 0 of 7 |
| Trains on your prompts | does not say | does not say |
| Keeps your requests | does not say | does not say |
| How you pay | does not say | does not say |
| Smallest top-up | does not say | does not say |
| Free tier | none listed | none listed |
| Where it runs | does not say | does not say |
Every model both sell
7 models, most widely sold first. Per M tokens, blended 3 in : 1 out.
| Model | DeepInfra | Groq |
|---|---|---|
| Qwen3.8 27BAlibaba Cloud · 1M context | $0.58 $0.15 in · $1.88 out | $1.60 $0.80 in · $4 out |
| gpt-oss-120bOpenAI · 131k context | 7¢ 3.7¢ in · $0.17 out | $0.26 $0.15 in · $0.60 out |
| Llama 3.3 70B InstructMeta · 131k context | $0.16 $0.10 in · $0.32 out | $0.64 $0.59 in · $0.79 out |
| MiniMax-M2.7MiniMax · 205k context | $0.44 $0.25 in · $1 out | $0.90 $0.60 in · $1.80 out |
| gpt-oss-20bOpenAI · 131k context | 5.8¢ 3¢ in · $0.14 out | $0.13 7.5¢ in · $0.30 out |
| Qwen3.6 27BAlibaba Cloud · 262k context | $1.04 $0.32 in · $3.20 out | $1.20 $0.60 in · $3 out |
| Llama 3.1 8B InstructMeta · 131k context | 2.5¢ 2¢ in · 4¢ out | 5.8¢ 5¢ in · 8¢ out |
How we count
DeepInfra is cheaper on 7 of the 7 models both sell, Groq on 0. Both are the same kind of provider, hosts running the same models, so the price is most of what separates them.
Each price is the cheapest that provider charges for the model, per M tokens, blended 3 in : 1 out. The smaller line under it is the price in and the price out.
A free tier is a limit rather than a price, and a price below the maker's own with nothing published to account for it is left out, so neither can be marked as the cheaper of the two.
Frequently asked
- Is DeepInfra cheaper than Groq?
- On the 7 models both sell, DeepInfra is cheaper on 7, Groq on 0. The widest gap is Llama 3.3 70B Instruct: $0.16 per million tokens against $0.64. Prices are per M tokens, blended 3 in : 1 out, and read on 31 Oct 2026.
- Which models do DeepInfra and Groq both sell?
- 7 of the models we track, led by Qwen3.8 27B, gpt-oss-120b, Llama 3.3 70B Instruct. DeepInfra sells 126 in all and Groq 9, so most of each catalog has no counterpart at the other.
- Does DeepInfra or Groq train on what you send it?
- DeepInfra does not say whether it trains on what you send it, and does not say how long it keeps requests. Groq does not say whether it trains on what you send it, and does not say how long it keeps requests. A provider that may train on prompts is often how a price below the maker's own is paid for.
- How current are these prices?
- Both catalogs are read every 3 days. DeepInfra was last read 31 Oct 2026 and Groq 31 Oct 2026; the whole index was refreshed 31 Oct 2026. Prices exclude tax, and a price below the maker's own with nothing published to account for it is left out of this table entirely.
Provider pages