GLM-5.3 on Run BiOS

GLM-5.3 is a large-scale reasoning model from Z.ai, built for complex software engineering and long-horizon agent tasks. It uses the same base model as GLM-5.2 — every gain comes from post-training. Compared with GLM-5.2, it is much better at complex coding and long-horizon tasks:

• Stronger coding. Z.ai describes GLM-5.3 as the most capable open-weights model for coding, with a 50% improvement over GLM-5.2 on its in-house Z.ai Code Bench, and open-source SOTA on public benchmarks including Terminal Bench 3.0 and Agents' Last Exam.
• Emergent cyber capability. As post-training scaled, cyber capability developed faster than Z.ai expected: GLM-5.3 is state of the art on CyberGym for vulnerability discovery, and its gains are largest further up the exploitation chain, where it more than doubles GLM-5.2 on exploitation benchmarks.

Architecture

GLM-5.3 shares GLM-5.2's base: a mixture-of-experts model with DeepSeek Sparse Attention (DSA). Published model configuration:

• Scale: 744B total parameters with 40B active per token (744B-A40B); BF16 weights.
• Context: a maximum position embedding of 1,048,576 tokens (1M context); text input and output.
• Configuration: 78 hidden layers (the first 3 dense, the remainder sparse MoE), hidden size 6144, 256 routed experts with 8 routed experts active per token plus 1 shared expert, MoE intermediate size 2048, dense-layer intermediate size 12288, 64 attention heads, MLA-style attention with Q LoRA rank 2048 and KV LoRA rank 512, vocabulary size 154,880.
• Speculative decoding: one multi-token-prediction (MTP) layer.

Thinking and Reasoning Effort

Per the official model card:

• GLM-5.3 supports controlling the thinking budget through the reasoning_effort parameter, which accepts three levels: low, high, and max. It defaults to max if not passed (or if set to any other value). To use low or high, pass them explicitly.
• For benchmark and leaderboard reproduction, Z.ai recommends keeping the default max.
• In the model's chat template, clear_thinking defaults to false; Z.ai recommends explicitly passing clear_thinking=true for chat scenarios.

Benchmarks

Full benchmark table as published in the official GLM-5.3 model card.

Benchmark	GLM-5.3	GLM-5.2	Kimi K3	DeepSeek-V4 Pro-0813	Qwen3.8-Max	Opus 4.8	Fable 5 (w/ fallback)	GPT-5.6 Sol
Terminal Bench 2.1	88.2	81.0	88.3	87.9	86.6	85.0	88.0	88.8
Terminal Bench 3.0	28.3	4.6	17.4	-	-	21.1	33.7	34.6
DeepSWE (v1.1)	66.9	46.2	67.5	62.7	56.6	58.0	69.7	72.7
NL2Repo	58.0	48.9	58.0	61.1	55.9	69.7	-	-
ProgramBench (Almost Solved)	19.0	9.5	17.5	-	10.5	15.5	33.0	23.0
FrontierSWE	78.1	67.5	-	-	-	66.5	88.2	-
SWE-Marathon (v1.1)	42.5	19.4	48.1	-	-	48.8	33.1	42.5
PostTrainBench	39.8	31.7	32.0	-	-	32.9	41.8	36.2
CyberGym	84.5	77.2	80.0	83.3	78.5	78.1	83.8	83.6
ExploitGym (2h / 6h)	105 / 130	29 / 39	36 / 70	-	14 / 26	80 / 120	181 / 247	216 / 293
ExploitBench	54.4	24.4	32.2	-	28.8	40.0	78.0	76.5
Toolathlon Verified	73.0	59.9	76.5	74.1	72.5	76.2	74.7	74.9
AutomationBench (v1.0.6)	48.2	26.2	46.7	43.2	39.8	41.0	46.2	45.8
Agents' Last Exam (ALE-CLI)	28.5	23.8	27.6	25.7	27.0	25.7	23.8	28.6
HLE w/ Tools	62.5	54.7	59.8	60.0	56.2	57.9	63.9	64.5
GDPval-AA v2	1769	1508	1682	1590	1739	1588	1743	1730

Evaluation Notes (as published by Z.ai)

• HLE w/ tools: sampling temperature=1.0, top_p=0.95, maximum generation length 163,840 tokens, maximum context length 300,000 tokens with a context management strategy; GPT-5.6-luna (medium) as the judge model.
• NL2Repo: temperature=1.0, top_p=1.0, max_new_tokens=64k under 1M context, with rule-based and LLM-based judgement to prevent malicious behaviors.
• DeepSWE: the mini-swe-agent harness with temperature=0.95, top_p=1.0, timeout=6h, and 400K context.
• Terminal-Bench 2.1: evaluated in Claude Code 2.1.207 with temperature=1.0, top_p=1, max_new_tokens=65536, and a 6-hour timeout.
• Terminal-Bench 3.0: the Claude Code 2.1.207 harness (reasoning effort max, 400K context, 128K maximum output), avg@3 over three rollouts per task, each in an isolated container built from the task's official image, capped at 600 agent turns with a 10-hour timeout; artifacts scored by the task's official separate verifier.
• Agents' Last Exam (CLI): the official evaluation protocol with the Claude Code harness (reasoning effort max, 1M context, 64K maximum output); each of the 105 tasks in an isolated container with its Task Card resources; scored by the official ALE evaluators.
• Toolathlon Verified: the official evaluation service, pass@1 averaged over 3 independent runs.
• AutomationBench: v1.0.6 with the null-type handling fix.
• GDPval-AA v2: evaluated by Artificial Analysis.
• CyberGym / ExploitGym / ExploitBench: Claude Code 2.1.207 at max reasoning effort with no web tools (temperature=1.0, top_p=1.0, max_new_tokens=128000), with a domain whitelist to prevent the agent from cheating; CyberGym is single-run Pass@1 over 1,507 tasks; ExploitGym is single-run Pass@1 on 869 tasks under 2-hour and 6-hour budgets rescaled by per-model tokens-per-second; ExploitBench averages coverage over 41 tasks across 3 revisions.
• FrontierSWE: 1M context length, max effort level, 128K maximum output tokens; dominance score as of 2026/08/14.
• PostTrainBench / SWE-Marathon: Claude Code 2.1.207 at max effort with a 1M-token context window and 128K maximum output tokens (see the official card for the anti-cheat adjustments documented for individual tasks).

License

GLM-5.3 is released under the GLM-5.3 License: a permissive grant (use, modify, distribute, sell, run, deploy, fine-tune, and create derivative works) with one condition beyond attribution — a licensee operating a Model-as-a-Service business whose aggregate revenue exceeds US$10 billion over any consecutive twelve months must pass Z.AI's security review before commercial use.

On Run BiOS

On the Run BiOS serverless catalog this model is served as glm-5.3 with its full 1,048,576-token context window, function calling, structured outputs, adjustable reasoning effort (exposed as a low / high / max ladder, matching the model's own three levels; the model's own default is max), and prompt caching for repeated prefixes.