2026-08-26 · Model Release

GLM-5.3
FLASH
FRONTIER COST

The first natively multimodal model of the GLM-5 series. 320B total parameters, 18B active. Claude Opus 4.8-class coding at one-tenth the price.

GLM-5.3-FLASH320B / 18B ACTIVE$0.045 / TASK1M CONTEXTAA INDEX 57OPUS 4.8-CLASS CODINGSERVED ON CHINESE AI CHIPS
0Total params
0Active
0AA Index v4.1
$0Per task
0×Cheaper
TL;DRRelease notes — the short version

GLM-5.3-Flash outperforms GLM-5.2 across benchmarks and real-world workloads at one-tenth the price, while approaching Claude Opus 4.8 on coding and agentic benchmarks.

For the first time, a hybrid architecture combines sparse and linear attention, sharply reducing long-context serving costs. Manifold-Constrained Hyper-Connections (mHC) improve scaling efficiency. A 30T-token multimodal pre-training corpus lets it produce more intelligence with less compute.

⚡ Stealth drop

Before release, we tested it anonymously as ox-alpha on OpenCode and OpenRouter. It quickly became the most popular model of the week — all traffic served on Chinese AI chips.

01

Competitive performance,
flash cost

GLM-5.3-Flash pushes the Pareto frontier of the Artificial Analysis Intelligence Index v4.1.1, scoring 57 at $0.045 per task — intelligence previously available only at roughly 10× the cost.

Across six coding and agentic benchmarks it consistently outperforms GLM-5.2: 63.4 vs 46.2 on DeepSWE v1.1, 48.8 vs 26.2 on AutomationBench — while approaching Claude Opus 4.8 overall.

On Z.ai Code Bench at max effort: 29.0 vs 29.5 — a near tie with Opus 4.8.

02

Architecture for
extreme efficiency

Despite a similar total parameter count (320B vs 355B), GLM-5.3-Flash nearly halves the activated parameters (18B vs 32B) and layers (45 vs 92) versus GLM-4.5.

Linear attention captures local dependencies through state modeling; sparse attention retrieves global context through a lightweight indexer. IndexPool compresses four indexer key vectors into one via weighted pooling.

Attention compute 3.0× smaller. KV cache 4.4× smaller. Lowest attention compute of all models compared.

BenchmarkGLM-4.5-BaseGLM-5-BaseDS-V4-Flash-BaseGLM-5.3-Flash-Base
Specs
Activated Params32B40B13B18B
Total Params355B744B284B320B
Base model results
MMLU86.188.388.588.1
BBH86.287.484.986.6
HellaSwag87.188.185.387.1
LiveCodeBench-Base28.134.429.937.6
SimpleQA30.036.031.233.5
Code lets the model build and change the world. Vision lets it enter the world people see and use.
— GLM-5.3-Flash research notes
03

Visual intelligence
in the coding loop

Visual coding expands what code can reach. For frontend, game development, and 3D simulation, the output is not code alone — it's an interface, an interaction, a world. Many failures only surface through rendering, interaction, playtesting.

Vision is natively integrated: the model decides when to observe and uses visual feedback to guide actions. Its pipelines demand self-visual judgment — interacting with environments, inspecting outputs, refining iteratively.

Validation beyond functional correctness — the rendered, interactive product.

04

Beyond coding —
your partner at work

Professional work is interpreting heterogeneous visual and structured information — documents, spreadsheets, presentations, dashboards. Rather than translating the environment into text, the model directly interprets the artifacts of the task, assesses its own output against visual context, and refines.

Stronger judgments of presentation quality and aesthetics — self-verification that sees what the user sees.

05

Serving at scale on
Chinese AI chips

A dedicated inference engine on top of SGLang — accelerated by the GLM-5.3-powered infrastructure agent. A feedback loop where the model helped optimize the system serving the model itself.

Intra-node tensor parallelism, ReplaySSM, W8A8 quantization, hybrid INT8/FP8/BF16 cache quantization, Layer Split — plus Encode–Prefill–Decode disaggregated architecture at cluster scale.

End-to-end serving improvement
= NVIDIAHardware efficiency & per-token cost
06

Full benchmark
results

BenchmarkGLM-5.3-FlashGLM-5.2DS-V4-VisionOpus 4.8GPT-5.6 TerraGemini 3.7 Flash
Coding
Terminal Bench 2.184.381.083.985.087.485.8
DeepSWE v1.163.446.259.358.069.665.3
NL2Repo56.348.957.769.7
Agentic
Toolathlon Verified78.459.975.976.274.9
AutomationBench v1.0.648.826.238.841.037.252.3
Agents' Last Exam26.320.427.327.028.0
HLE w/ Tools55.354.755.157.9
GDPval-AA v2177315041675158215711527
Vision
OfficeQA Pro62.457.948.9
CharXiv Reasoning89.480.489.988.088.7
Chartography w/ Tools78.064.375.068.065.0
BabyVision53.435.146.861.670.9
MVbench77.869.467.175.082.2
MMVU80.572.767.475.882.3
Footnotes & evaluation protocols

HLE w/ tools: temperature=1.0, top_p=0.95, max 163,840 tokens, context 300K. Judge: GPT-5.6-luna (medium).

DeepSWE: mini-swe-agent harness, temperature=0.95, timeout 6h, 400K context.

Terminal-Bench 2.1: Claude Code 2.1.207, temperature=1.0, max 6h timeout.

Toolathlon: official evaluation service, pass@1 averaged over 3 runs.

BabyVision: shorter side ≥1.5K pixels.

FRONTIER
INTELLIGENCE.
FLASH COST.

All GLM Coding Plan users get it with 3× the usable quota of GLM-5.3. Weights are public on HuggingFace for local deployment with SGLang, vLLM and TokenSpeed.

Try GLM-5.3-Flash Download weights ↗