DeepInfra raises $107M Series B to scale the inference cloud — read the announcement

Kimi K2 0905 is a state-of-the-art large language model developed by Moonshot AI, representing a significant advancement in open-weight AI capabilities. This Mixture-of-Experts (MoE) model features 1 trillion total parameters with 32 billion activated parameters per forward pass, making it highly efficient while maintaining frontier-level performance. The model supports a 256k token context window and excels at agentic coding intelligence, tool calling, frontend development, and long-horizon autonomous tasks.
Trained using the innovative MuonClip optimizer on 15.5 trillion tokens, Kimi K2 0905 delivers exceptional performance across coding, math, and reasoning benchmarks. It is specifically designed for tool use, reasoning, and autonomous problem-solving — making it well suited for developers building AI agents and complex automation workflows.
Kimi K2 0905 is now available across multiple inference providers. This analysis breaks down which one delivers the best performance, lowest cost, and fastest response times for your use case.
| Provider | Why Notable | Speed (t/s) | TTFT (s) | Blended ($/1M) | E2E (s) | Context | JSON | Func |
|---|---|---|---|---|---|---|---|---|
| DeepInfra | Best overall value: lowest price + lowest latency with solid throughput | 77.7 | 0.53s | $0.80 | 6.96s | 131k | Yes | Yes |
| Groq | Best for throughput-intensive workloads: fastest generation speed | 202.1 | 1.26s | $1.50 | 3.73s | 262k | Yes | Yes |
| Fireworks | Mid-pack performance; higher cost than DeepInfra | 42.5 | 1.44s | $1.20 | 13.22s | 262k | Yes | Yes |
| Novita | Budget alternative; slowest speed and highest latency | 27.5 | 1.99s | $1.07 | 20.18s | 262k | Yes | Yes |
Based on benchmarks across 4 tracked providers, DeepInfra is the recommended API for production-scale Kimi K2 0905 deployment. It offers the lowest latency (0.53s TTFT), the lowest blended price ($0.80/1M tokens), and solid throughput (77.7 t/s). The only scenario where an alternative makes sense is when maximum generation speed is the primary requirement — in which case Groq’s 202.1 t/s throughput justifies its premium pricing.
DeepInfra delivers the optimal balance of performance and cost for Kimi K2 0905, making it the best choice for the vast majority of production deployments.
DeepInfra’s sub-second latency (0.53s) makes it ideal for interactive applications where responsiveness directly impacts user experience. Combined with its industry-leading pricing, it offers the best total cost of ownership for production workloads. The slightly smaller context window (131k vs 262k) may be a consideration for extremely long-context applications, but for the vast majority of use cases, DeepInfra delivers unmatched value.
Groq’s custom LPU (Language Processing Unit) architecture delivers unparalleled generation speed, making it the go-to choice for throughput-intensive applications.
Groq’s 202.1 t/s output speed is nearly 3x faster than the next competitor, making it exceptional for batch processing, real-time streaming applications, or scenarios where generation time is the critical bottleneck. However, this performance comes at a premium — at $1.50/1M blended, it costs nearly double DeepInfra’s rate. Choose Groq when raw speed matters more than cost optimisation.
Fireworks offers a middle-ground option with reliable performance but doesn’t lead in any single metric.
Fireworks provides consistent, reliable service with full feature support and a larger context window than DeepInfra. It is a reasonable choice for enterprises already integrated into the Fireworks ecosystem, though DeepInfra offers better value across all performance metrics for new deployments.
Novita offers lower pricing than Groq and Fireworks, but with significant performance trade-offs that limit its practical applicability.
Novita’s pricing falls between DeepInfra and Fireworks, but its performance lags significantly behind all three. A 20-second end-to-end time for 500 tokens makes it unsuitable for latency-sensitive applications. DeepInfra still offers better pricing with vastly superior performance, making Novita difficult to recommend for most use cases.
For teams planning future projects, Moonshot AI’s newer Kimi K2.5 model — released in January 2026 — represents a significant evolution with several key upgrades:
If your use case involves vision-based inputs, multi-agent orchestration, or advanced UI generation, Kimi K2.5 is worth evaluating for your next project.
For Kimi K2 0905 deployments, DeepInfra is the recommended provider for most use cases. Its combination of the lowest latency (0.53s TTFT), the lowest blended price ($0.80/1M tokens), solid throughput (77.7 t/s), and full JSON Mode and Function Calling support makes it the optimal choice for production applications.
GLM-4.7-Flash API Benchmarks: Latency, Throughput & Cost<p>About GLM-4.7-Flash GLM-4.7-Flash is Z.AI’s open-weights reasoning model released in January 2026. Built on a Mixture-of-Experts (MoE) Transformer architecture, it features 30 billion total parameters with only ~3 billion active per inference — making it exceptionally efficient for its capability class. The model is designed as a lightweight, cost-effective alternative to Z.AI’s flagship GLM-4.7, optimized […]</p>
GLM-5.1 on DeepInfra: Z.AI’s Agentic Engineering Model<p>Z.AI’s GLM-5.1 scores 58.4 on SWE-Bench Pro — ahead of both Claude Opus 4.6 (57.3) and GPT-5.4 (57.7) on real-world software engineering tasks. It’s the direct successor to GLM-5, designed for agentic engineering: long-horizon coding tasks, terminal operations, and repository-level work. The core design premise is that previous models, including GLM-5, tend to plateau after […]</p>
NVIDIA Nemotron 3 Super on DeepInfra: 120B MoE Model<p>NVIDIA’s Nemotron 3 Super runs 120 billion parameters while activating only 12 billion per token — a ratio that makes a real difference when orchestrating multiple agents in parallel. It’s built on a novel architecture called LatentMoE, a hybrid of Mamba-2, Mixture-of-Experts, and Attention layers designed from the ground up for agentic, reasoning, and long-context […]</p>
© 2026 DeepInfra. All rights reserved.