Z.ai-reported result
Terminal-Bench 3.0
GLM-5.328.3
GLM-5.24.6
Z.ai reports 28.3 for GLM 5.3 and 4.6 for GLM 5.2 on Terminal-Bench 3.0, the largest reported jump in the launch material.
GLM 5.3 Is Officially Announced
GLM 5.3
GLM 5.3 is the new Zhipu AI (Z.ai) flagship large language model: 1M-token context, 128K maximum output, and top open-source coding benchmarks. Chat with GLM 5.3 free on glm5.app, and track release date, pricing, and open-source news here.
Create with GLM 5
1 Credits
What Is the GLM 5.3 Model?
GLM 5.3 is the new flagship large language model from Zhipu AI (Z.ai), announced in August 2026. It builds on the GLM 5.2 base with large-scale post-training on real long-horizon engineering workflows.
01
The new Z.ai flagship
GLM 5.3 is Z.ai newest flagship base model with a 1M-token context window, 128K maximum output, and text-in, text-out modality. It is the GLM family model that follows GLM 5.2.
02
Same base, far more training
GLM 5.3 uses the same base as GLM 5.2 and adds tens of times more long-horizon post-training across real compute clusters, storage systems, internal docs, and codebases, which Z.ai says improves coding feel by about 50%.
03
An honest release tracker
The GLM 5.3 model API is coming soon, so this page separates what Z.ai has confirmed (specs and benchmarks) from what is still pending (API ID, pricing, weights, and license).
GLM 5.3 Release Status: Confirmed Facts and Pending Details
Z.ai has confirmed GLM 5.3 specs and benchmarks. API access, pricing, and open weights are still pending official announcements.
Confirmed by Z.ai
GLM 5.3 verified facts
- Context window
- 1M tokens
- Maximum output
- 128K tokens
- Modality
- Text in / text out
- GLM Coding Plan
- Live
- Model API
- Coming soon
Official GLM 5.3 model page lists a 1M-token context window.
Official maximum output tokens per response.
GLM 5.3 is a text base model; no vision or multimodal input is documented.
Z.ai Coding Plan subscription has fully rolled out GLM 5.3.
Z.ai says the GLM 5.3 model API will go live shortly and has not published a firm date.
Pending official details
Tracked expectations
- GLM 5.3 pricing
- Unannounced
- Open weights
- Expected, unconfirmed
- Parameters
- Unpublished
- Official model ID
- Expected glm-5.3
Reference: GLM 5.2 API is about $1.40 input / $4.40 output per 1M tokens.
GLM 5.2 ships under MIT on Hugging Face; GLM 5.3 weights and license are not published yet.
Z.ai has not published GLM 5.3 parameter totals.
glm5.app maps the GLM 5.3 chat option to its current backend until the official API goes live.
Status is checked against Z.ai public communications as of August 2026. Details not marked as confirmed should be treated as unverified until Z.ai announces them.
Benchmarks / 02
GLM 5.3 benchmarks vs GLM 5.2: use them as a starting point
Z.ai reports large benchmark gains for GLM 5.3, but a score only applies to the published harness and date. Reproduce a task that resembles your own workflow before making a production decision.
Z.ai-reported result
DeepSWE v1.1
GLM-5.366.9
GLM-5.246.2
Z.ai reports 66.9 for GLM 5.3 and 46.2 for GLM 5.2 on DeepSWE v1.1, a long-horizon software engineering benchmark.
Z.ai-reported result
Agents' Last Exam
GLM-5.328.5
GLM-5.223.8
Z.ai reports 28.5 for GLM 5.3 and 23.8 for GLM 5.2 on Agents' Last Exam, covering agentic knowledge and reasoning.
Z.ai-reported result
CyberGym
GLM-5.384.5
GPT-5.6 Sol83.6
Z.ai reports 84.5 for GLM 5.3 on CyberGym, slightly above GPT-5.6 Sol at 83.6. It is a published result, not a guarantee for your security workflow.
Capabilities / 03
GLM 5.3 capabilities
GLM 5.3 supports the standard developer stack for agentic and long-horizon work, with a 1M-token context window to hold project-scale context.
01
Thinking modes
Multiple thinking modes let you balance reasoning depth against latency, matching the task at hand.
02
Function calling
Strong tool-calling support for external integrations, with your application validating and executing every requested action.
03
Streaming output
Real-time streaming responses for interactive chat and agent loops.
04
Context caching
Smart caching optimizes long conversations and repeated context, cutting cost on long-horizon sessions.
05
Structured output
JSON and other structured output formats make system integration predictable.
06
MCP support
GLM 5.3 can call external MCP tools and data sources, extending agent workflows beyond chat.
Release timeline / 04
GLM 5.3 release timeline and news
The milestones that led to GLM 5.3, and what still needs to land. This timeline is updated whenever Z.ai publishes official GLM 5.3 news.
01
Jun 16, 2026 — GLM 5.2 launches
GLM 5.2 launched for long-horizon tasks with a 1M-token context window and effort-level control.
02
Aug 2026 — GLM 5.3 announced
Z.ai published the official GLM 5.3 model page and benchmark results, naming it the new flagship built on the GLM 5.2 base.
03
Aug 2026 — Coding Plan goes live
The GLM Coding Plan subscription fully rolled out GLM 5.3 for coding tools like Claude Code, Cline, and OpenCode.
04
Next — GLM 5.3 model API
Z.ai says the GLM 5.3 API is coming soon. The official model ID, limits, and pricing will be updated here when announced.
05
Next — open weights and license
GLM 5.3 weights and license details are expected to follow the GLM 5.2 MIT precedent on Hugging Face, pending official confirmation.
06
Where GLM 5.3 news breaks first
Follow the Z.ai blog, the zai-org GitHub repository, and the official bigmodel docs for the first GLM 5.3 API announcements.
Who it is for / 05
Who uses GLM 5.3?
GLM 5.3 is aimed at teams doing long-horizon engineering, agentic workflows, and security review with project-scale context.
01
Application developers
Developers building coding assistants and IDE workflows benefit from the reported 50% coding feel improvement over GLM 5.2 and the 1M-token context for whole-repository work.
02
Agent builders
Agent teams use thinking modes, function calling, streaming, and MCP to build long-running autonomous workflows with clear stop conditions.
03
Security researchers
GLM 5.3 shows strength in white-box code review and vulnerability discovery, with CyberGym at 84.5, useful as an assistant layer that experts still validate.
04
Long-horizon engineering teams
Teams working across tens of thousands of lines of code and dozens of files use GLM 5.3 to plan refactors, fix bugs, and deliver projects with minimal hand-holding.
05
API and platform teams
Platform teams track GLM 5.3 API pricing and limits to budget for context caching and structured output before the official API goes live.
06
Open-source community
Self-hosters watch for GLM 5.3 open weights on Hugging Face to plan hardware, quantization, and license review, following the GLM 5.2 MIT precedent.
How to use GLM 5.3 / 06
How to use GLM 5.3
GLM 5.3 is available in glm5.app Chat today, with the official model API expected to follow.
01
01
Start a GLM 5.3 chat
Open GLM 5.3 Chat on glm5.app, describe the problem, constraints, and desired outcome, then refine the response with your own code and acceptance criteria.
02
02
Subscribe to the GLM Coding Plan
For coding-tool integrations today, the Z.ai Coding Plan already serves GLM 5.3 in tools like Claude Code, Cline, and OpenCode.
03
03
Integrate when the API goes live
When Z.ai releases the official GLM 5.3 model API, this page will link the model ID, limits, and pricing. Until then, use the GLM 5.3 chat option with the current glm5.app backend model.
Integration specimen
Start with the GLM 5.3 API
The official GLM 5.3 model API is coming soon from Z.ai. glm5.app exposes an OpenAI-compatible Chat Completions endpoint today, and the GLM 5.3 model ID will be added here when Z.ai publishes the official API. Keep API keys on the server, start with a narrow request, and record the expected response shape before scaling.
Open API documentationRequest previewcurl
curl https://glm5.app/api/v1/chat/completions \
-H "Authorization: Bearer $GLM5_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "glm-5.3",
"messages": [{
"role": "user",
"content": "Plan this refactor and list the rollback steps."
}]
}'GLM 5.3 FAQ
Clear GLM 5.3 answers about release date, pricing, benchmarks, Hugging Face, parameters, weights, the coding plan, and GLM 5.3 vs 5.2.
What is GLM 5.3?
GLM 5.3 is the new flagship large language model from Zhipu AI (Z.ai), announced in August 2026. It keeps the same base as GLM 5.2 and adds large-scale post-training on long-horizon tasks, with a 1M-token context window and 128K maximum output. It is a text-in, text-out model.
When is the GLM 5.3 release date?
Z.ai published the official GLM 5.3 model page and benchmarks in August 2026. The GLM Coding Plan already serves GLM 5.3, and the GLM 5.3 model API is expected to go live shortly; Z.ai has not published a firm API date yet.
What is GLM 5.3 pricing?
Z.ai has not published GLM 5.3 API pricing yet. For reference, GLM 5.2 costs about $1.40 per million input tokens and $4.40 per million output tokens on the official API. glm5.app offers free GLM 5.3 chat credits, and GLM 5.3 pricing will be updated here once Z.ai announces it.
What are the GLM 5.3 benchmarks?
Z.ai reports Terminal-Bench 3.0 at 28.3 (up from 4.6 for GLM 5.2), DeepSWE v1.1 at 66.9 (up from 46.2), Agents Last Exam at 28.5 (up from 23.8), and CyberGym at 84.5. Benchmarks only apply to the published harness, so evaluate GLM 5.3 on your own workflow before deciding.
Is GLM 5.3 on Hugging Face?
GLM 5.3 weights are not on Hugging Face yet. Z.ai open-sourced GLM 5.2 under the MIT license, so a GLM 5.3 release on Hugging Face is widely expected, but it has not been published as of August 2026.
How many parameters does GLM 5.3 have?
Z.ai has not published GLM 5.3 parameter counts. The official model page lists a 1M-token context window and 128K maximum output, and GLM 5.2 documents 1M context. Parameter totals are expected to follow the GLM 5 family MoE architecture, subject to official confirmation.
Will GLM 5.3 weights be open source?
GLM 5.3 weights are expected to be open source under the MIT license, following the GLM 5.2 precedent on Hugging Face. Z.ai has not confirmed the GLM 5.3 license yet, so treat open-weights details as unconfirmed until the release.
What is the GLM 5.3 coding plan?
The GLM Coding Plan, Z.ai subscription for AI coding tools like Claude Code, Cline, and OpenCode, has fully rolled out GLM 5.3. It is a subscription product rather than an open API, so integration terms differ from the GLM 5.3 model API that is still coming.
GLM 5.3 vs GLM 5.2: what changed?
GLM 5.3 keeps the GLM 5.2 base and adds much larger long-horizon post-training, which Z.ai says improves coding feel by about 50%. Reported gains include Terminal-Bench 3.0 from 4.6 to 28.3 and DeepSWE v1.1 from 46.2 to 66.9, plus new cybersecurity capability in white-box code review.
How can I use GLM 5.3 now?
You can chat with GLM 5.3 free in glm5.app Chat today. The GLM Coding Plan also serves GLM 5.3. The public GLM 5.3 model API is expected to go live soon; this page will update with the official model ID and limits when Z.ai announces them.
Is glm5.3 the same as GLM 5.3?
Yes — glm5.3, GLM-5.3, and GLM 5.3 are spelling variants of the same model. The official API model ID is expected to be glm-5.3, and glm5.app maps the GLM 5.3 chat option to the current backend model until the official API goes live.
Start here
Start with GLM 5.3 today
Chat with GLM 5.3 free on glm5.app, and track the official API, pricing, and open-source news here.