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

Updated November 2026.
GLM-5.3 is Z.ai’s reasoning model for long-horizon coding and agentic tasks, with a 1M-token context window. Serving it at scale raises practical questions about cost per token, output speed, and how long a request waits before the first answer token arrives.
This guide compares seven providers that offer GLM-5.3. Speed and latency figures come from Artificial Analysis benchmarks on a 10,000-token prompt, and pricing and features come from each provider’s published documentation. Because GLM-5.3 always reasons before answering, time to first answer token includes thinking time, so it runs well above the time to first chunk.
A quick reference by priority:
| Provider | Best For |
|---|---|
| DeepInfra | The lowest blended price in this guide ($0.45 per 1M tokens, level with GMI Cloud), plus JSON mode, function calling, and private endpoints. |
| Z.ai | Direct first-party access and GLM Coding Plan subscriptions. |
| GMI Cloud | Teams that want serverless inference and bare-metal GPU rental on one platform. |
| FriendliAI | High-throughput agent workloads backed by a published 99.99% uptime SLA. |
| Telnyx | Teams already on Telnyx’s network that want inference on the same infrastructure. |
| Inco | The fastest output speed and lowest latency, at the highest blended price. |
| DigitalOcean | Teams already on DigitalOcean that want GLM-5.3 beside 70+ other models. |
Pricing as of 2026-11-16.
DeepInfra is the best overall pick when cost per task drives the decision. Artificial Analysis lists it at $0.45 per 1M tokens blended, half Z.ai’s first-party $0.90 and sixth-lowest of the 30 GLM-5.3 endpoints it tracks.
The cached input rate matters most for agentic and RAG workloads, which resend large prompt prefixes on every turn. At $0.125 per 1M tokens, cached input costs about 78% less than the promotional input rate. The full rate card, including Flex, is on DeepInfra’s pricing page, and the GLM-5.3 API reference covers reasoning_effort and the OpenAI-compatible endpoint. Measured output speed sits mid-field, so applications that need the fastest streaming should weigh Inco (FAST).
Z.ai is the model’s creator and offers first-party API access. The same API underpins its GLM Coding Plan subscriptions.
Z.ai ships new releases first and supports the coding plans directly. Its per-token rates run above DeepInfra’s current listing. GLM-5.3 shares its base model with GLM-5.2, which DeepInfra covers in its GLM-5.2 model overview.
GMI Cloud combines serverless inference with bare-metal GPU rental on one platform.
GMI suits teams that train or fine-tune on rented GPUs and want to serve the result from the same platform. Its measured GLM-5.3 throughput sits near the bottom of the field, second-lowest of 22 providers.
FriendliAI serves GLM-5.3 through its Model APIs, with dedicated endpoints available for reserved capacity.
FriendliAI publishes its uptime and speed claims platform-wide, so validate them on your own GLM-5.3 workload. The Artificial Analysis figures above are the independent reference point.
Telnyx runs GLM-5.3 on GPUs it owns and operates, reachable through an OpenAI-compatible Chat Completions API.
Telnyx suits teams already using its voice or messaging network, since inference stays on the same infrastructure. Independent speed data is not yet available.
Inco leads Artificial Analysis’s speed and latency rankings for GLM-5.3 through its Inco (FAST) tier.
Inco (FAST) suits real-time applications where response speed outweighs cost. Budget-sensitive workloads will find it the most expensive option.
DigitalOcean offers GLM-5.3 through Serverless Inference, alongside more than 70 other models on one endpoint.
DigitalOcean fits teams already running on its cloud who want GLM-5.3 next to other models under one bill. Latency-sensitive workloads should look at faster providers.
The right GLM-5.3 provider depends on which constraint binds first: price, speed, or platform fit. Measure latency as time to first answer token, since thinking time dominates the wait.
DeepInfra is the best overall pick when cost per task decides the choice. It pairs the lowest blended price in this guide with JSON mode, function calling, cached input at $0.125 per 1M tokens, and private endpoints. Workloads that need the fastest streaming should benchmark Inco (FAST) first.
Lighter workloads can run on GLM-5.3-Flash at a lower per-token price. Teams comparing generations can review DeepInfra’s GLM-5.2 pricing and cost comparison, and qualifying startups can apply to DeepStart for up to 1 billion free tokens.
Kimi K3 vs Claude Opus 4.8 vs GPT-5.6 Sol: Practical AI Model Comparison<p>What’s changed since this comparison first ran Two of the three models compared here have since been succeeded, and GPT-5.6 Sol’s pricing has changed materially. Three of the strongest models available on DeepInfra right now don’t separate cleanly by capability tier. Kimi K3 (available via DeepInfra), Claude Opus 4.8, and GPT-5.6 Sol all score within […]</p>
MiMo-V2.5 Is Now Available on DeepInfra<p>Xiaomi’s MiMo-V2.5 collapses what used to require two separate models — frontier agentic capability and native multimodal understanding — into one. Previously, MiMo-V2-Pro handled agentic and coding tasks while MiMo-V2-Omni covered visual and audio inputs; MiMo-V2.5 replaces both. It handles text, images, video, and audio natively, extends context to 1 million tokens, and scores 71.8 […]</p>
Best MiMo-V2.5 API Providers Ranked<p>Executive Summary: Selecting the right API provider for Xiaomi’s MiMo-V2.5 is critical for optimizing production workflows. Based on the benchmark research, DeepInfra is the best provider for raw speed and low latency (130+ tokens/second), while Xiaomi’s first-party API is the most cost-effective, offering unmatched prompt caching discounts. This guide breaks down the model’s MoE architecture […]</p>
© 2026 DeepInfra. All rights reserved.