GLM 5.3 lands on Amazon Bedrock: 753B MoE with stronger coding and standout cybersecurity scores
AWS ML Blog · rss · 2026-10-06
Z.ai's GLM 5.3 is now available on Amazon Bedrock for eligible enterprise customers.
Model highlights:
- 753B-parameter MoE optimized for coding and long-horizon agentic workflows (multi-hour tasks, repo-wide refactoring)
- Z.ai reports a 50% coding improvement over GLM 5.2 on internal benchmarks, plus competitive results on DeepSWE, Terminal Bench 3.0, and FrontierSWE
- Leading CyberGym score of 84.5 at release, with notable cybersecurity capabilities
Bedrock integration:
- OpenAI-compatible Responses / Chat Completions APIs plus Bedrock Invoke / Converse
- Implicit prompt caching by default; explicit caching via promptcachebreakpoint markers (≥1024 tokens each) to cut latency and input costs for agentic workloads
- US and Global cross-region inference profiles (us.zai.glm-5.3, global.zai.glm-5.3) and Flex / Standard / Priority service tiers
The post includes full Python examples (short-lived credentials, explicit caching) and a demo workflow using the open-source AI pentesting agent Strix for authorized security testing.
More from Infra
- Reflection Ships Apache 2.0 Model With Tech Report; Analyst Estimates Pre-training MFU at Just ~12% — eliebakouch · 2026-10-06
- Anthropic moves Claude Cowork fully to the cloud, dropping the battery-hungry local VM — Simon Willison · 2026-10-06
- ik_llama.cpp MoE fork: expert residency and hybrid execution give 20-30% speedup on 6GB GPUs — IceFog72 · 2026-10-06
- AI tools quietly ate 44 GB of his disk, so he built free cleaner Sparewise — PossibilityKind3028 · 2026-10-06
- Nebius hikes RAM 41% and GPUs up to 21% as the chip shortage hits cloud price lists — tengyanAI · 2026-10-06
- GitHub Actions goes down again, breaking CI pipelines — generativist · 2026-10-06