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

Updated November 2026.
Deploying the GLM-5.2 (max) Mixture-of-Experts model — 753B total parameters with roughly 40B active per token and a 1M context window — requires infrastructure that separates production-grade API providers from the rest. This guide breaks down the top providers by throughput, latency, pricing, and quantization architecture.
Newer: GLM-5.3. Z.ai has released GLM-5.3, and Artificial Analysis now recommends it over GLM-5.2. GLM-5.2 remains available on DeepInfra. See the GLM-5.3 API provider benchmarks.
| Provider | Why it’s a strong option for GLM-5.2 (max) | Best-fit use cases | What to confirm before choosing |
|---|---|---|---|
| DeepInfra | Serves GLM-5.2 as zai-org/GLM-5.2 (set reasoning_effort to “max” for GLM-5.2 (max)) at $0.5625 / $1.80 / $0.105 per 1M on promotion (list $0.75 / $2.40 / $0.14), with Flex ($0.45 / $1.44 / $0.084) and Priority tiers, FP4, a 1M context and up to 163,840 output tokens. 0.0% error rate on Artificial Analysis. | Long-context applications (1M tokens), cost-sensitive agent and batch workloads (Flex), reasoning-heavy workloads where you still want open-weights flexibility | Rate limits for your account, and whether Flex’s queueing (up to 10 minutes before an HTTP 429) fits your latency budget |
| Other providers (comparison) | Artificial Analysis reports pricing and performance as first-party API (if available) or median across providers when first-party isn’t available; provider-to-provider variance can materially change cost and latency. | Cost-sensitive deployments (this model is pricey vs peers), latency-sensitive apps (TTFT matters), workloads that can exploit caching | Check per-provider price, TTFT, throughput, caching policy, and availability |
| API Provider | Output Speed (t/s) | Time to First Answer Token* | Error Rate | Blended Price (7:2:1, per 1M) | Input / Output (per 1M) | Quantization / Precision | Best For |
|---|---|---|---|---|---|---|---|
| Multiverse Computing | 111.0 | 19.84 s | 0.0% | $0.10 | $0.17 / $0.54 | Not stated | Lowest price |
| DeepInfra | 81.1 | 34.97 s | 0.0% | $0.37 | $0.5625 / $1.80 | FP4 | Overall value |
| CoreWeave | 214.1 | 10.84 s | 2.3% | $0.46 | $0.62 / $2.42 | Not stated | Speed at a low price |
| GMI | Not measured | Not measured | — | $0.59 | $0.923 / $2.903 | FP8 | Budget |
| Baseten | 101.6 | 33.60 s | 2.9% | $0.82 | $1.40 / $4.40 | Not stated | Standard endpoint |
| Mistral | 214.2 | 8.18 s | 4.1% | $0.82 | $1.40 / $4.40 | Not stated | Shortest wait to first answer |
| Nebius | 144.1 | 27.13 s | 0.0% | $0.82 | $1.40 / $4.40 | FP4 | Balanced |
| SiliconFlow | 102.0 | 27.88 s | 0.0% | $0.85 | $1.302 / $4.092 | FP8 | Balanced |
| FriendliAI | 113.6 | 22.26 s | 0.0% | $0.90 | $1.40 / $4.40 | Not stated | Balanced |
| Novita | 86.5 | 31.59 s | 0.0% | $0.90 | $1.40 / $4.40 | FP8 | General use |
| Parasail | 179.5 | 22.04 s | 10.2% | $0.90 | $1.40 / $4.40 | NVFP4 | Throughput |
| Databricks | 222.8 | 13.11 s | 0.0% | $0.90 | $1.40 / $4.40 | Not stated | Enterprise |
| Z.ai (first-party) | 66.2 | 29.64 s | 0.0% | $0.90 | $1.40 / $4.40 | Not stated | Baseline |
| Baseten (FAST) | 266.8 | 11.63 s | 0.0% | $1.23 | $2.10 / $6.60 | Not stated | Raw speed |
| Scaleway | 148.1 | 13.30 s | 0.0% | $2.52 | $2.09 / $6.38 | Not stated | EU hosting |
*Note: For reasoning models like GLM-5.2 (max), time to first answer token includes the model’s internal “thinking” time before the final answer starts.
GLM-5.2 was Z.ai’s flagship model before GLM-5.3, built for coding, reasoning, and tool-driven agentic workloads. Released in June 2026, it succeeds GLM-5.1 in the GLM-5 family and represents a significant evolution from the original GLM-5 (744B parameters) released in February 2026.
Z.ai — formerly Zhipu AI — became a publicly traded foundation model company with its Hong Kong IPO in January 2026. The company, founded in 2019 as a spin-off from Tsinghua University, has established itself as a leader in open-source AI research with a consistent release cadence.
As measured by Artificial Analysis in June 2026, on an earlier version of its Intelligence Index, GLM-5.2 (max) scored 51, ahead of MiniMax-M3 (44), DeepSeek V4 Pro (44), and Kimi K2.6 (43). Artificial Analysis has since deprecated GLM-5.2 and shows no score for it on the current index. The model was reportedly trained on Huawei Ascend chips using the MindSpore framework — a notable detail given Z.ai’s placement on the U.S. Entity List, which restricts access to NVIDIA H100/H200 GPUs.
DeepInfra is the overall recommended API provider for GLM-5.2 (max). Serving a large 753B parameter MoE model is notoriously difficult, but DeepInfra leverages FP4 (4-bit floating point) quantization to achieve an efficient deployment.
Why FP4 matters: NVIDIA’s Blackwell architecture (B200 GPUs) features native FP4 tensor cores that enable hardware-accelerated FP4 compute. FP4 quantization can theoretically achieve meaningful speedup compared to BF16 inference while delivering substantial memory reduction, with accuracy recovery generally improving on larger MoE architectures like GLM-5.2 (max). This allows DeepInfra to support the model’s full 1,048,576-token context window.
At $0.5625 per 1M input tokens and $1.80 per 1M output tokens on current promotion (list $0.75 / $2.40; cached input $0.105), DeepInfra has the second-lowest blended price Artificial Analysis measures for GLM-5.2 ($0.37, after Multiverse Computing) and the lowest listed output price on OpenRouter’s GLM-5.2 list, with a 0.0% error rate. Flex ($0.45 / $1.44 / $0.084) and Priority ($0.84375 / $2.70 / $0.1575) tiers are available, and the endpoint accepts up to 163,840 output tokens. Pricing as of 2026-11-17. Run zai-org/GLM-5.2 with reasoning_effort set to “max” for GLM-5.2 (max); the API reference lists every parameter.
If response speed is your product, test CoreWeave (214.1 t/s at $0.62 / $2.42) or Databricks; for cost-sensitive, batch and agent workloads, DeepInfra’s per-token rates compound.
GMI lists GLM-5.2 at $0.923 / $2.903 per 1M tokens ($0.59 blended), fourth-lowest on Artificial Analysis, which has no current speed figures for its endpoint. GMI uses FP8 quantization, which reduces memory bandwidth requirements compared to standard FP16 deployments while retaining output quality.
GMI’s pricing structure makes it attractive for batch processing or high-volume, automated agentic tasks where millisecond latency is not the primary constraint.
Baseten runs two GLM-5.2 endpoints on Artificial Analysis: Baseten (FAST) at 266.8 t/s and 11.63 seconds to first answer token ($2.10 / $6.60 per 1M), and a standard endpoint at 101.6 t/s and 33.60 seconds ($1.40 / $4.40). Serving a model that activates 40B parameters per token requires robust GPU orchestration, and Baseten’s infrastructure handles this efficiently.
Baseten is a reasonable endpoint for developers who need sustained, high throughput for real-time coding assistants or complex multi-step reasoning applications.
Databricks offers a performant endpoint for enterprise users, achieving an output speed of 222.8 t/s and 13.11 seconds to first answer token at $1.40 / $4.40, with a 0.0% error rate.
For teams already embedded in the Databricks ecosystem, this provider offers a way to integrate GLM-5.2 (max)’s reasoning capabilities and context window into existing data pipelines and software engineering workflows.
Z.ai’s native API provides the baseline experience for GLM-5.2 (max). Pricing is explicit: $1.40 per 1M input tokens, $4.40 per 1M output tokens, and a discounted $0.26 per 1M cached tokens (resulting in a blended price of roughly $0.90).
The first-party API provides Anthropic-compatible endpoints (https://api.z.ai/api/coding/paas/v4), making it straightforward to integrate into existing tools like Claude Code or Cline. Artificial Analysis measures Z.ai’s endpoint at 66.2 t/s; its main draw is day-zero feature support and native integration of the model’s “high” and “max” reasoning effort modes.
Understanding quantization is useful when selecting a provider for large MoE models like GLM-5.2 (max):
Providers serving GLM-5.2 (max) must also optimize for KV cache scaling to handle the 1M-token context window, typically using techniques like continuous batching and efficient memory management via inference frameworks such as vLLM or TensorRT-LLM.
Deploying a large Mixture of Experts model like GLM-5.2 (max) requires API providers to push the boundaries of hardware optimization and memory bandwidth. The model’s 1M-token context window, 40B active parameters per token, and intensive reasoning capabilities demand infrastructure that can balance throughput, latency, and cost.
By using FP4 quantization on modern GPU architectures, DeepInfra (deepinfra.com) balances the model’s large context window with a $0.37 blended price on Artificial Analysis at current promotional pricing. Whether you are building long-horizon coding agents or complex reasoning systems, DeepInfra provides a technically capable infrastructure option for GLM-5.2 (max).
Best Kimi K3.1 Pro API Providers (2026): Price, Tiers, Speed<p>Summary of the Best Kimi K3.1 Pro Providers Provider Best For DeepInfra Lowest price of the providers in this guide ($1.90 / $9.50), Flex and Priority tiers, 0% measured error rate, private endpoints Moonshot AI (Kimi API) First-party access, plus the Highspeed variant and Moonshot’s own coding and work apps DigitalOcean Fast responses at Moonshot’s […]</p>
GLM-5.3 Is Now Available on DeepInfra<p>GLM-5.3 shares its base model with GLM-5.2, so every performance gain comes from post-training. Z.ai scaled the reinforcement learning stack it already had and released the result on August 14, 2026. GLM-5.3 is a reasoning model built for complex software engineering and long-horizon agentic tasks. On Z.ai’s internal Code Bench, it improves 50% over GLM-5.2, […]</p>
© 2026 DeepInfra. All rights reserved.