1M-Token Lossless Context
GLM-5.2 keeps high retrieval accuracy across the full 1,000,000-token window — about 5x the prior GLM-5 limit — letting you load an entire monorepo or long spec into one GLM-5.2 API call.
GLM 5.2
GLM-5.2 is Z.ai's (formerly Zhipu AI) open-weight, 753B-parameter Mixture-of-Experts model with a lossless 1M-token context window, trained for long-horizon agentic coding and repository-wide refactoring. Call the GLM-5.2 API on GPTProto at $1.26 / $3.96 per 1M tokens — 10% under Z.ai's list rate, with one key and one balance across 200+ models.
API Usage Examples
curl --request POST "https://gptproto.com/v1/chat/completions" \
--header "Authorization: Bearer $GPTPROTO_API_KEY" \
--header "Content-Type: application/json" \
--data '{
"model": "glm-5.2",
"messages": [
{
"role": "user",
"content": "Hello"
}
]
}'GLM 5.2 pricing
Chat, coding agents & document work. Priced per 1M tokens — input, cached input and output are billed separately. GPTProto is 10% below official rates.
| Scenario | Z-AI list | OpenRouter | GPTProto | You save / mo |
|---|---|---|---|---|
| Personal10M tokens / mo (4.8M cached) | $14.53 | $15.33 | $13.08 | −$1.45≈ $17.43 / yr |
| Team100M tokens / mo (48M cached) | $145.28 | $153.27 | $130.75 | −$14.53≈ $174.34 / yr |
| Business500M tokens / mo (240M cached) | $726.40 | $766.35 | $653.76 | −$72.64≈ $871.68 / yr |
Save $0.00 / month
Annual savings ≈ $0.00 · pay as you go
Related Models
All Models
GLM 5.2Current
Z-AI · 1.05M context · $1.26/M input · $3.96/M output
Bytedance · 262K context · $0.0971/M input · $0.9714/M output
Bytedance · 262K context · $0.0971/M input · $0.9714/M output
Model
Price: GPTProto
vs Official
vs OpenRouter
Context
Stability
GLM 5.2Current
Z-AI
$1.26 / $3.96per 1M tokens
-10%
-15%
1.05M
GLM-5.2 API
What sets the GLM-5.2 API apart — a 1M-token lossless context, IndexShare sparse attention, MIT open weights, and selectable High/Max reasoning effort.
Agentic-RL Training
GLM-5.2 went through months of reinforcement learning on long-horizon coding tasks, so it sustains autonomous, multi-step engineering work across hundreds of iterations without the early plateau of earlier GLM-5 releases.
IndexShare Architecture
A sparse-attention design that reuses one indexer across every four layers, cutting per-token attention compute by up to 2.9x at the full 1M context — paired with multi-token prediction for faster GLM-5.2 API inference.
Permissive MIT License
GLM-5.2's weights ship under the OSI-approved MIT license as open source — free to self-host, fine-tune, and use commercially with no per-seat fees, a rarity at this capability tier.
What is GLM-5.2?
GLM-5.2 is the June 2026 flagship in Z.ai's (formerly Zhipu AI) GLM-5 series — an open-weight Mixture-of-Experts model built specifically for long-horizon coding agents rather than general chat. It pairs a usable 1M-token context with native tool calling, MCP support, and structured JSON output, and ships under an MIT open-source license you can self-host. On GPTProto, the GLM-5.2 API is metered pay-as-you-go access with no regional sign-up barriers and one balance shared across 200+ models.
| Spec | GLM-5.2 |
|---|---|
| Developer | Z.ai (formerly Zhipu AI) |
| Architecture | Mixture-of-Experts, ~753B total / ~40B active per token |
| Context window | 1,000,000 tokens (lossless) |
| Max output | 131,072 tokens (~128K) |
| Modality | Text in, text out |
| Reasoning | Selectable effort — High / Max |
| Tool use | Native function calling, MCP, structured JSON |
| License | MIT (open weights) |
| Released | June 13, 2026 |
| GPTProto price | $1.26 in / $3.96 out per 1M tokens (10% under Z.ai list) |
| API model string | glm-5.2 (provider slug z-ai, full 1M context) |
GLM-5.2 vs GLM-5 and GLM-5.1
All three GLM-5 releases are live on GPTProto, so moving up a version is just a model-string swap — same key, same balance. The jump to 5.2 is mostly about context size and long-task stability:
| GLM-5 | GLM-5.1 | GLM-5.2 | |
|---|---|---|---|
| Released | Feb 2026 | Apr 2026 | Jun 2026 |
| Context window | ~200K-class | ~200K-class | 1,000,000 (≈5x) |
| SWE-bench Pro* | — | 58.4 | 62.1 |
| Long-context attention | standard | standard | IndexShare (−2.9x compute) |
| Speculative decoding | — | — | MTP (+~20% accepted tokens) |
| GPTProto output / 1M | $2.88 | $3.96 | $3.96 |
The practical takeaway: GLM-5.2 lists at the same $3.96 output rate as GLM-5.1 on GPTProto but gives you roughly 5x the context and a measurable bump in long-horizon coding — so for repository-scale work there's little reason to stay on 5.1.
GLM-5.2 vs GPT-5.5
GPT-5.5 is also on GPTProto, so this is a real swap, not a marketing comparison. On Z.ai's reported long-horizon coding benchmarks, GLM-5.2 edges GPT-5.5 on most agentic tasks while costing far less per token:
| Benchmark (vendor-reported, Z.ai) | GLM-5.2 | GPT-5.5 |
|---|---|---|
| SWE-bench Pro | 62.1 | 58.6 |
| FrontierSWE (long-horizon) | 74.4% | 72.6% |
| MCP-Atlas (tool use) | 77.0 | 75.3 |
| Humanity's Last Exam (w/ tools) | 54.7 | 52.2 |
| Terminal-Bench 2.1 | 81.0 | 84.0 |
| GPTProto price / 1M (in / out) | $1.26 / $3.96 | $4 / $24 |
On GPTProto the GLM-5.2 API runs about 3x cheaper on input and 6x cheaper on output than GPT-5.5 ($1.26/$3.96 vs $4/$24 per 1M). GPT-5.5 still leads on raw terminal tasks, so it isn't a clean sweep — but for agentic, tool-driven, multi-step coding the benchmark edge favors GLM-5.2 at a fraction of the spend. On the independent Artificial Analysis Intelligence Index v4.1, GLM-5.2 scores 51, competitive at the open frontier though not ahead of every closed model on general tasks.
Switching from the official Z.ai API
If you're already calling GLM-5.2 on Z.ai's API, moving to GPTProto is a drop-in: keep your request body, change the base URL and key. You get the same glm-5.2 model with no separate Z.ai account, no regional sign-up friction, and one balance that also covers GPT-5.5, Claude, Gemini, and 200+ other models. The exact endpoint and model string are in the Quick Start below.
Glm 5.2 FAQ: Technical Details & Usage
Get answers about glm 5.2 performance, context limits, and local deployment via Zhipu AI's open-weight framework.
How does glm 5.2 handle massive context windows?
GLM-5.2 runs a full 1,000,000-token context — roughly 5x the prior GLM-5 window — and holds high retrieval accuracy across it via its IndexShare sparse-attention design, which cuts per-token compute by up to 2.9x at max length. In practice you can load an entire monorepo or long spec into one GLM-5.2 API call.
Is glm 5.2 truly open-weight for commercial use?
Yes. GLM-5.2's weights are released under the MIT license, so you can download, self-host, fine-tune, and ship commercial products with no per-seat fees. If you'd rather not run GPUs, the GLM-5.2 API on GPTProto gives you the same model as metered access.
What is the glm 5.2 reasoning_effort parameter?
GLM-5.2 exposes a selectable reasoning-effort setting with two levels — High and Max. High favors speed and efficiency; Max spends more tokens on deeper, more thorough reasoning for hard coding or analysis. You set it per request.
How does glm 5.2 compare to Claude Opus 4.8?
On Z.ai-reported coding benchmarks GLM-5.2 lands just behind Claude Opus 4.8 — SWE-bench Pro 62.1 vs 69.2, FrontierSWE 74.4% vs 75.1%, MCP-Atlas 77.0 vs 77.8 — at a fraction of the price. Both run on GPTProto under one balance, so you can route routine tasks to GLM-5.2 and reserve Opus 4.8 for the hardest ones.
Does GLM-5.2 support native tool calling?
Yes — GLM-5.2 has first-class function calling, multi-turn tool use, MCP integration, and structured JSON output, which is why it's built for autonomous coding agents. The GLM-5.2 API on GPTProto is OpenAI-compatible, so existing tool-calling code works with a base-URL change.
How much does the GLM-5.2 API cost on GPTProto?
$1.26 per 1M input tokens and $3.96 per 1M output tokens — about 10% under Z.ai's $1.40 / $4.40 list rate. Billing is metered pay-as-you-go from one balance; see the Pricing section for worked cost examples.
Is there a free GLM-5.2 or GLM-5 API?
There's no permanently free hosted tier. Because the weights are MIT-licensed, GLM-5.2 and GLM-5 are free to run if you self-host on your own GPUs. For hosted access without buying hardware, the GLM-5.2 API on GPTProto is metered — you pay only for the tokens you use, with no regional sign-up barriers and no subscription.
Does GPTProto offer the GLM Coding Plan?
No — the GLM Coding Plan is Z.ai's flat subscription. GPTProto provides metered GLM-5.2 API access instead: no monthly commitment, no peak-hour quota throttling, and one key and balance that also cover GPT-5.5, Claude, Gemini, and 200+ other models.
Related Articles
Guides, comparisons, and updates related to this model.
All Articles
What Is GLM 5.2? Open-Weight Coding at 1/6 the Price
GLM 5.2 is Z.ai's open-weight, MIT-licensed coding model with a 1M-token context. See its features, benchmarks vs Claude Opus 4.8 and GPT-5.5, pricing, and how to run it.
Best Text-to-Image APIs in 2026: 7 Models Ranked by Quality and Price
The 7 best text-to-image APIs in 2026, ranked by Arena Elo and real price — GPT Image 2, Nano Banana, Seedream 5.0. Call them all with one API key.
Nano Banana Pro vs Nano Banana 2: Which Gemini Image Model Should You Use in 2026?
Nano Banana 2 costs half of Nano Banana Pro and scores higher on the Image Arena. See when each Gemini image model wins—plus one-API code to run both.
How to Illustrate a Children's Book with AI (Print-Ready and Character-Consistent for ~$1)
Build a print-ready (300 DPI), character-consistent AI picture book — same character on every page, KDP-ready files — for about $1 in API cost. Full code.