DeepInfra published a comparative ranking of API providers for Xiaomi's MiMo-V2.5 reasoning model on July 1, 2026, positioning its own hosted offering as the best choice for raw speed and low latency at 130+ tokens/second. The model is an MoE with 310B total / 15B active parameters, trained on 48 trillion tokens, suppo The article benchmarks DeepInfra alongside other vendors, reporting MiMo-V2.5 median output throughput of 87.2 t/s and TTFT of 2.76s, with an estimated Artificial Analysis Intelligence Index of 40. The piece gives developers concrete guidance on provider selection across speed, latency, and cost dimensions, and emphasi

DeepInfra published a benchmark analysis on April 30, 2026 comparing six inference providers for DeepSeek V4 Pro (Max) across latency, throughput, and cost. The model is a 1.6T-total / 49B-active MoE with a 1M-token context window, pre-trained on 32T+ tokens using the Muon optimizer, and introduces a hybrid attention a The article highlights Fireworks as fastest at 167.1 t/s output and Together.ai as lowest TTFT at 0.9s, while DeepInfra is benchmarked using FP4 quantization. The piece is positioned as a multi-provider competitive review that includes DeepInfra's own offering, providing technical signal on how DeepInfra's FP4 deployme

Deep Infra: AI infrastructure provider offering API access to hosted open-source AI models for developers, focused on low-latency inference at scale.

DeepInfra's benchmark post (2026-04-03) compares four Kimi K2 0905 inference providers on latency, throughput, and blended price, recommending DeepInfra overall with the lowest blended price at $0.80 per 1M tokens and the lowest time-to-first-token at 0.53s. Groq leads throughput at 202.1 tokens/second and posts the fa The post surfaces one notable DeepInfra limitation: its Kimi K2 0905 context window tops out at 131k tokens versus 262k on the other three providers, which is a concrete trade-off for long-context workloads. The piece is now roughly 3.5 months old relative to the 2026-07-26 review date, and figures are vendor self-publ

PitchBook's company profile confirms that Deep Infra closed a $107 million Series B on April 16, 2026, bringing its total early-stage venture funding activity to a Series A (completed December 6, 2024) and this Series B, both reported as generating revenue. The profile lists 26 investors across its funding history and According to the profile, Deep Infra operates an inference platform that exposes models for token classification, text-to-image, object detection, automatic speech recognition, text generation, embeddings, and similar tasks, aimed at accelerating AI-powered application development. The Series B funding event is the rel