DeepInfra
A cloud platform providing serverless GPU inference services. It hosts a variety of open-weight models and is known for low-cost and FP8 endpoints.
Timeline (last 90 days)
According to DeepSeek's disclosures, a single cluster running the DeepInfra stack can serve 155 trillion tokens per day.
DeepInfra offers a batch-invariant mode that forces mathematical results to be identical every time; this mode is off by default because computing the same way every time is slower.
DeepInfra's kernels are written for NVIDIA data center Blackwell chips (such as the B200) and cannot be used out of the box on devices using different Blackwell chips, such as the DGX Spark.
The DeepSeek v4.1 Flash model can already run on conventional inference engines such as vLLM, and DeepInfra was initially built on vLLM as well.
Qwen/Qwen3-235B-A22B-Instruct-2507 ranges from a low of $0.09/M (DeepInfra) to a high of $3/M (Crusoe) across 3 providers, a spread of +3233.3%, the largest price spread case in the article.
Llama 3.3 70B Instruct on DeepInfra's FP8 serverless endpoint has an input price as low as $0.12/1M.
All Facts (6)
According to DeepSeek's disclosures, a single cluster running the DeepInfra stack can serve 155 trillion tokens per day.
60%UnverifiedDeepInfra offers a batch-invariant mode that forces mathematical results to be identical every time; this mode is off by default because computing the same way every time is slower.
50%UnverifiedDeepInfra's kernels are written for NVIDIA data center Blackwell chips (such as the B200) and cannot be used out of the box on devices using different Blackwell chips, such as the DGX Spark.
50%UnverifiedThe DeepSeek v4.1 Flash model can already run on conventional inference engines such as vLLM, and DeepInfra was initially built on vLLM as well.
50%UnverifiedQwen/Qwen3-235B-A22B-Instruct-2507 ranges from a low of $0.09/M (DeepInfra) to a high of $3/M (Crusoe) across 3 providers, a spread of +3233.3%, the largest price spread case in the article.
50%UnverifiedLlama 3.3 70B Instruct on DeepInfra's FP8 serverless endpoint has an input price as low as $0.12/1M.
50%