GLM-5.3 cyber eval: 75% CVE rediscovery beats GPT-5.6, costs 40% less
joshua_saxe · x · 2026-08-15
A third-party security evaluation of GLM-5.3 shows it rediscovered 75% of CVEs at pass@3, 7 points higher than GPT-5.6-Terra, while costing 40% less to run. At pass@1, it surfaces 60.4% of CVEs on average, the highest for an open-source model. It also reports fewer false positives than DeepSeek v4 models, making its output more trustworthy. The evaluator calls GLM-5.3 the strongest and most consistent OSS model for cyber tasks.
More from Models
- Unredacted Anthropic filing exposes internal data-quality debate months before Opus 3 cutoff — zacharynado · 2026-10-15
- Google says teams across the company are using Gemini 4 Argon on its hardest engineering problems — CurieuxExplorer · 2026-10-15
- Perplexity's new embedding models: index with a 9B model, search with 0.6B — antoine_chaffin · 2026-10-15
- 797GB of n-gram counts in CPU RAM let 1B models match 57% more compute — PangWeiKoh · 2026-10-15
- Leak: Moonshot's Kimi flagship 'x4' done with pre-training, K3.1 in internal testing — lyraxana · 2026-10-15
- Unsloth ships GGUF quants for Nemotron 3.5: Bolt-74B runs in 24GB RAM — danielhanchen · 2026-10-15