DeepInfra raises GPU-hour rates and DeepSeek-V3.1 token price
DeepInfra lifted dedicated GPU rates 16-32% (H100 to .20, B200 to .69, B300 to .89/GPU-hr) and raised DeepSeek-V3.1 to /bin/bash.25//bin/bash.95 per 1M tokens, a notable reversal for a vendor known for cutting prices.
Custom-LLM GPUs: H100 .79, H200 .19, B200 .79, B300 .20/GPU-hr; on-demand 1xB200 .79/hr. DeepSeek-V3.1 /bin/bash.21 in / /bin/bash.79 out per 1M.
Custom-LLM GPUs: H100 .20, H200 .69, B200 .69, B300 .89/GPU-hr (A100 flat at /bin/bash.89); on-demand 1xB200 .69/hr, 8xB200 9.52/hr. DeepSeek-V3.1 /bin/bash.25 in / /bin/bash.95 out per 1M.
DeepInfra, whose reputation in cost-sensitive communities was built on relentless public price cuts, raised its dedicated GPU-hour rates for the first time across the tracked span. Custom-LLM rates moved H100 .79 to .20 (+23%), H200 .19 to .69 (+23%), B200 .79 to .69 (+32%), and B300 .20 to .89 (+16%); only the A100 held at /bin/bash.89/GPU-hour. On-demand B200 instances moved in lockstep — a single card went from .79 to .69/hr and an 8-card box from 2.32 to 9.52/hr.
On the per-token side, the popular DeepSeek-V3.1 rose to /bin/bash.25 in / /bin/bash.95 out per 1M (from /bin/bash.21 / /bin/bash.79), while Llama-3.1-8B-Turbo output edged down to /bin/bash.03. The model catalog also expanded (DeepSeek-V4-Flash /bin/bash.09//bin/bash.18, gemini-3.x, gemma-4, claude-fable-5 0/0, claude-sonnet-5 /0). DeepCluster reserved pricing (.98-.99/GPU-hr), the 0-0,000 usage tiers, and the 1.5x Priority service tier were unchanged.
DeepInfra raises its dedicated GPU-hour rates for the first time in the tracked span: custom-LLM H100 $1.79→$2.20, H200 $2.19→$2.69, B200 $2.79→$3.69, B300 $4.20→$4.89 per GPU-hour (A100 holds at $0.89); on-demand B200 instances move in lockstep (1×B200 $2.79→$3.69/hr, 8×B200 $22.32→$29.52/hr). DeepSeek-V3.1 per-token rises to $0.25 in / $0.95 out (from $0.21 / $0.79) and Llama-3.1-8B-Turbo output drops to $0.03. The model catalog expands (DeepSeek-V4-Flash $0.09/$0.18, gemini-3.x, gemma-4, claude-fable-5 $10/$50, claude-sonnet-5). DeepCluster ($1.98–$2.99/GPU-hr), usage tiers ($20–$10,000), and the Priority service tier (1.5×) are unchanged (source: deepinfra.com/pricing & /gpu-instances 2026-07-14).