Skip to content
← Models
Open weightsZhipu AI/10 trustVerified

GLM-5.3

Zhipu's open-weight coding flagship — a 744B MoE that pushed the frontier of autonomous cyber capability.

Visit official page →

About

GLM-5.3 is Zhipu's current flagship, a ~744B-total / ~40B-active MoE released August 2026 that gains its improvements over GLM-5.2 entirely from scaled post-training. Zhipu positions it as the strongest open-weight coding model, and it demonstrated emergent autonomous cyber capabilities during training.

The smaller GLM-5.3-Flash (320B total / 18B active, MIT-licensed) is the model behind the "Ox Alpha" stealth release that topped OpenRouter's leaderboard in August 2026. GLM-5.3 joined Amazon Bedrock under a revenue-sharing deal on October 5, 2026.

Specs

Context window

1,000,000 tokens (128,000 max output)

Pricing

$0.15 input / $0.50 output per 1M tokens ($0.03 cached input) — verified figures are for GLM-5.3-Flash; full GLM-5.3 API pricing not verified.

Free quota

None verified.

API access

Zhipu BigModel API, Amazon Bedrock (added Oct 5, 2026), third-party gateways

Weights

Open weights

Released

August 14, 2026

Benchmarks

Zhipu-reported:

  • Terminal-Bench 3.0: 28.3
  • DeepSWE v1.1: 66.9
  • Agents' Last Exam (ALE-CLI): 28.5

Independent-adjacent: Anthropic reported GLM-5.3 developing end-to-end exploits in 50 of 410 ExploitBench attempts, near Claude Mythos Preview's 56.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

chatcodevisionreasoningautomation

Trust score — 9/10

/10 trustVerified

Strong coding/agentic capability with open weights and a notable Bedrock partnership signaling enterprise validation. But: all headline benchmarks are vendor-claimed, the license is bespoke, and Anthropic's report of near-parity with Claude on autonomous exploits — plus easily-stripped refusals and a reported incident of its ZCode tool uploading user repos to Alibaba Cloud without consent — raises safety and data-handling questions developers should weigh.

How the trust score is built

Full methodology →
Adoption · stars/downloads, log-scaled vs peers
10.0/10
Maintenance · recency, releases, issue load
9.0/10
Openness · license + docs
7.2/10
Manual review · sanity read + red-flag check
6.0/10
  • · HF top match: zai-org/GLM-5.3-Flash (community upload counts as proxy where official repo is gated)
  • · HF downloads 6,489,462 (peer-normalized)
  • · released 3 days ago
  • · manual review: safety concerns (prompt injection, weaponization)

Sources