Baseten vs DeepInfra
Same workloads, both price lists, refreshed daily. On shared line items today: Baseten is cheaper on 0 of 18 shared models (input or output price), DeepInfra on 17.
Baseten
Production inference platform with Truss packaging, optimized serving engines and enterprise-grade autoscaling. Strong for teams shipping custom models with SLAs.
- Optimized model serving (TensorRT-LLM)
- Enterprise autoscaling and observability
DeepInfra
Consistently the price floor for open-weight models: Llama, Qwen, DeepSeek at rock-bottom per-token rates with an OpenAI-compatible API and per-request GPU billing options..
- Among the cheapest open-model tokens anywhere
- OpenAI-compatible API
- Dedicated GPU deployments
Shared models, priced by both
List prices refreshed 2026-10-20 ยท input $/1M tokens| Model | Baseten in/out | DeepInfra in/out | Cheaper input |
|---|---|---|---|
| GPT-OSS 120B | $0.100 / $0.500 | $0.037 / $0.170 | DeepInfra |
| DeepSeek V4 Flash | $0.130 / $0.260 | $0.090 / $0.180 | DeepInfra |
| DeepSeek V4.1 Flash | $0.300 / $1.20 | $0.140 / $0.420 | DeepInfra |
| DeepSeek-V3.1 | $0.500 / $1.50 | $0.250 / $0.950 | DeepInfra |
| Inkling Small | $0.500 / $1.20 | $0.450 / $1.20 | DeepInfra |
| GLM-4.7 | $0.600 / $2.20 | $0.400 / $1.75 | DeepInfra |
| Kimi K2.5 | $0.600 / $3.00 | $0.450 / $2.25 | DeepInfra |
| Kimi K2 | $0.600 / $2.50 | $0.500 / $2.00 | DeepInfra |
| GLM-4.6 | $0.600 / $2.20 | $0.500 / $2.00 | DeepInfra |
| Nemotron 3 Ultra | $0.600 / $2.40 | $0.500 / $2.20 | DeepInfra |
| DeepSeek V3 | $0.770 / $0.770 | $0.320 / $0.890 | DeepInfra |
| Kimi K2.7 Code | $0.950 / $4.00 | $0.680 / $3.40 | DeepInfra |
| Kimi K2.6 | $0.950 / $4.00 | $0.750 / $3.50 | DeepInfra |
| GLM-5 | $0.950 / $3.15 | $0.600 / $2.08 | DeepInfra |
| Inkling | $1.00 / $4.05 | $0.950 / $4.05 | DeepInfra |
| GLM-5.2 | $1.40 / $4.40 | $0.750 / $2.40 | DeepInfra |
| DeepSeek V4 Pro | $1.74 / $3.48 | $1.30 / $2.60 | DeepInfra |
| Kimi K3 | $3.00 / $15.00 | $2.85 / $14.25 | DeepInfra |
Cheaper-on-count is a tally of listed prices, not a quality verdict: throughput, reliability and quantization differ between providers. Disclosure.