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

Updated November 2026.
GLM-5.2 is an open-source large language model from Z.AI, engineered for long-horizon coding, agentic, and reasoning tasks. Designed for complex reasoning, advanced software engineering, and large-scale data processing, GLM-5.2 introduces a massive 1,048,576-token context window alongside significant architectural innovations.
Hosted on the DeepInfra platform, GLM-5.2 provides developers with a high-performance, OpenAI-compatible interface. Whether you are building agentic workflows, analyzing entire codebases, or processing lengthy documents, GLM-5.2 offers the stability and intelligence required for next-generation AI applications.
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 model overview and integration guide.
GLM-5.2 was released on June 13, 2026, succeeding GLM-5.1 in the GLM-5 family. Unlike previous iterations, this model is engineered to maintain output quality and stability even when the 1M-token context is fully utilized, allowing for the seamless processing of large datasets and complex, multi-file repositories in a single prompt.
IndexShare and MTP: To support this context window efficiently, Z-AI introduced IndexShare, a mechanism that reuses the same indexer across every four sparse attention layers, resulting in a reported 2.9x reduction in per-token FLOPs at maximum context length. An upgraded Multi-Token Prediction (MTP) layer also optimizes speculative decoding, increasing token acceptance length by up to 20% for faster, more cost-effective generation.
Flexible Reasoning: Z.ai’s own API exposes GLM-5.2’s thinking depth as a “Flexible Effort” system with High and Max modes, and Z.ai recommends Max for complex, multi-step tasks. On DeepInfra, you set it with the reasoning_effort parameter: none, minimal, low, medium, high, xhigh or max. Setting it to "none" turns thinking off for fast, simple calls.
Open Access: GLM-5.2 is released under the MIT license, allowing unrestricted commercial use, modification, and self-hosting.
GLM-5.2 demonstrates strong performance across industry-standard evaluations, frequently rivaling or approaching proprietary models such as GPT-5.5 and Claude Opus 4.8.
| Category | Benchmark | GLM-5.2 | GLM-5.1 | Qwen3.7-Max | GPT-5.5 | Claude Opus 4.8 |
|---|---|---|---|---|---|---|
| Reasoning | GPQA-Diamond | 91.2 | 86.2 | 90.0 | 93.6 | 93.6 |
| Math | AIME 2026 | 99.2 | 95.3 | 97.0 | 98.3 | 95.7 |
| IMOAnswerBench | 91.0 | 83.8 | 90.0 | — | 83.5 | |
| Coding | SWE-bench Pro | 62.1 | 58.4 | 60.6 | 58.6 | 69.2 |
| FrontierSWE | 74.4 | 30.5 | — | 72.6 | 75.1 | |
| Agentic | MCP-Atlas | 76.8 | 71.8 | 76.4 | 75.3 | 77.8 |
Key Highlights
GLM-5.2 is accessible via DeepInfra’s OpenAI-compatible API, making integration straightforward for developers familiar with standard LLM tooling.
Use your DeepInfra API key in the Authorization header:
Authorization: Bearer <YOUR_DEEPINFRA_API_KEY>
curl -X POST https://api.deepinfra.com/v1/openai/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $DEEPINFRA_API_KEY" \
-d '{
"model": "zai-org/GLM-5.2",
"messages": [
{
"role": "user",
"content": "Explain the concept of speculative decoding in 2 sentences."
}
],
"temperature": 0.7| Parameter | Type | Description |
|---|---|---|
| model | String | Use “zai-org/GLM-5.2”. |
| messages | Array | Conversation history objects. |
| response_format | Object | Set to {“type”: “json_object”} for structured JSON. |
| tools | Array | Definitions for function calling. |
| temperature | Float | Controls randomness (0.0 to 2.0). |
| reasoning_effort | String | Thinking depth: “none”, “minimal”, “low”, “medium”, “high”, “xhigh” or “max”. “none” disables thinking. |
| max_tokens | Integer | Maximum tokens to generate in the response, up to 163,840. Defaults apply if unset. |
DeepInfra offers pay-per-token pricing for GLM-5.2, with Standard, Flex and Priority tiers. Standard is currently 25% off the list price. Pricing as of 2026-11-17; the GLM-5.2 model page always shows current rates.
| Tier | Input | Cached Input | Output |
|---|---|---|---|
| Standard (25% promotional discount) | $0.5625 / 1M tokens | $0.105 / 1M tokens | $1.80 / 1M tokens |
| Standard, list price | $0.75 / 1M tokens | $0.14 / 1M tokens | $2.40 / 1M tokens |
| Flex (0.8×) | $0.45 / 1M tokens | $0.084 / 1M tokens | $1.44 / 1M tokens |
| Priority (1.5×) | $0.84375 / 1M tokens | $0.1575 / 1M tokens | $2.70 / 1M tokens |
Flex runs at 0.8× the Standard rate for workloads that can tolerate variable latency, such as batch jobs and evaluations. The Priority Tier, at 1.5× the Standard rate, is designed for workloads requiring higher priority and faster processing. Flex and Priority multipliers apply to the current promotional Standard rate.
Among the GLM-5.2 providers Artificial Analysis tracks, DeepInfra’s Standard price is the second lowest, after Multiverse Computing, and Artificial Analysis measured a 0.0% error rate on DeepInfra’s endpoint. For the wider field, including SaaS tools, see our GLM-5.2 SaaS tools and API providers guide.
While standard API access is usage-based, users with high-throughput requirements can deploy Private Endpoints via the DeepInfra Dashboard for dedicated capacity.
GLM-5.2 combines a massive 1M-token context window with strong reasoning and coding capabilities, supported by architectural innovations like IndexShare and a flexible reasoning system. It provides developers with the efficiency and power needed for complex agentic and long-horizon tasks.
To begin building with GLM-5.2, visit the DeepInfra Dashboard to generate your API key and explore private deployment options, and see the GLM-5.2 API reference for every parameter.
Seed Anchoring and Parameter Tweaking with SDXL Turbo: Create Stunning Cubist ArtIn this blog post, we're going to explore how to create stunning cubist art using SDXL Turbo using some advanced image generation techniques.
Build a Streaming Chat Backend in 10 Minutes<p>When large language models move from demos into real systems, expectations change. The goal is no longer to produce clever text, but to deliver predictable latency, responsive behavior, and reliable infrastructure characteristics. In chat-based systems, especially, how fast a response starts often matters more than how fast it finishes. This is where token streaming becomes […]</p>
Unleashing the Potential of AI for Exceptional Gaming ExperiencesGaming companies are constantly in search of ways to enhance player experiences and achieve
extraordinary outcomes. Recent research indicates that investments in player experience (PX)
can result in substantial returns on investment (ROI). By prioritizing PX and harnessing
the capabilities of AI...© 2026 DeepInfra. All rights reserved.