GLM-5.3 Flash Cuts Costs by 17x without Sacrificing Much Performance
Researchers at Z.ai have unveiled GLM-5.3 Flash, a cost-optimized version of their flagship model GLM-5.3. According to a study on DeepSWE, a benchmark for software engineering tasks, GLM-5.3 Flash reduces rollout expenses by 17 times while retaining 94% of task coverage.
In the tests, GLM-5.3 Flash delivered 264 solves per $100 at $0.24 per rollout, far outpacing the 17 solves achieved by GLM-5.3 at $3.99 per rollout. The full model holds a 5.6-point lead in pass@1 accuracy (69.0% vs. 63.4%), but this gap narrows to just 2.6 points at pass@4 (87.6% vs. 85.0%).
The performance profile of GLM-5.3 Flash has been reshaped by distillation, rather than scaling it down uniformly. It improved results in specific domains such as concurrency (+8 points), Python (+5), and data modeling (+4), while ceding ground in JavaScript-heavy and reasoning-intensive tasks.
Despite the tradeoffs, the economics heavily favor GLM-5.3 Flash for throughput-driven workloads due to its lower cost and faster completion times (26 minutes on average compared to 35 minutes for GLM-5.3).