GLM 5.3

GLM 5.3 is a post-training revision built on the same base model as GLM 5.2, meaning all of its reported gains come from later training rather than changes to the foundation architecture. The model is positioned as an open-weights release aimed at complex coding and long-horizon agent workflows, and the release notes describe it as the most capable open-weights model for coding within the Z.ai lineup, with a roughly fifty percent improvement over GLM 5.2 on the in-house Z.ai Code Bench. On public coding benchmarks it reaches open-source state-of-the-art on Terminal Bench 3.0 and on Agents' Last Exam, and it posts competitive results against much bigger frontier systems on Terminal Bench 2.1, DeepSWE v1.1, NL2Repo, ProgramBench, FrontierSWE, and SWE-Marathon v1.1.

An unexpected outcome of scaling the post-training pipeline is a sharp jump in cyber capability, with the model described as state of the art on CyberGym for vulnerability discovery and showing its largest relative gains further along the exploitation chain, where it more than doubles the previous version on exploitation benchmarks. Practically, GLM 5.3 fits teams that want a locally deployable, open-weights model for sustained coding agents, multi-step software tasks, and security research, while accepting that headline numbers on the newest Terminal Bench tier remain behind a few larger proprietary systems. It is delivered through the Fireworks AI model key shown on this page and pairs the very large context window with reasoning, tool calling, structured output, and temperature control for building reliable agent loops.

Fireworks AIaccounts/fireworks/models/glm-5p3glm

Open provider page Provider docs

Quick Info

Provider: Fireworks AI
Model key: accounts/fireworks/models/glm-5p3
Release date: Aug 14, 2026
Last updated: Sep 7, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $1.40
Output token cost: $4.40

Limits

Output tokens: 262,144 tokens
Context window: 1,048,573 tokens

OpenCode

Model variants

Priority

Input

$1.75

per 1M tokens

Output

$5.50

per 1M tokens

Cache Read

$0.325

per 1M tokens

Transparent token rates

Compare GLM 5.3 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about GLM 5.3

Fireworks AI

Official source

Fireworks AI lists Z.ai’s GLM-5.3 as Ready under the model path `accounts/fireworks/models/glm-5p3`, with creation dated August 28, 2026. The page says the model uses the same base as GLM-5.2 and that its gains come from post-training focused on complex coding and long-horizon tasks. The model card confirms serverless access at $1.40 per million input tokens, $0.26 per million cached-input tokens, and $4.40 per million output tokens. It also lists LoRA fine-tuning, on-demand dedicated-GPU deployment, function calling, a 1040k-token context window, a 743B mixture-of-experts model, and no image-input

Fireworks AI

Coverage

Zhipu AI released GLM-5.3 on August 14, 2026, with open weights expected to follow about two weeks later (around August 28) on Zhipu's Hugging Face organization (zai-org) after an extensive risk review. The article frames this window as preparation time for infrastructure teams: Zhipu's internal evaluations put coding For self-hosting sizing, the article cites Z.ai's docs for the GLM-5 family architecture: Mixture of Experts, 744B total parameters, around 40B active per pass, with a 200K context (note: this conflicts with the 1M-token context cited via Z.ai's GLM-5.3 docs in the InferenceX candidate). At BF16, weights alone run near

Fireworks AI

CoverageBenchmark

GLM-5.3, announced on August 14, 2026 (per Yicai Global and Z.ai's blog/docs as cited on this page), is a post-training-only release from Z.ai (Zhipu AI) that reuses the unchanged GLM-5.2 base model — every gain comes from post-training. Z.ai frames it as "Frontier Coding with Emergent Cyber Capabilities," claiming a 5 The same Z.ai-sourced material on the page positions GLM-5.3 alongside its predecessor GLM-5.2 (introduced June 16, 2026, as Z.ai's flagship long-horizon model with a solid 1M-token context, IndexShare architecture, improved MTP speculative-decoding layer, and MIT-licensed open weights). Z.ai claims GLM-5.2 is the high