Key Info
A user benchmarked the GLM 5.3 model offered through OpenCode's Go plan at $10 for $500 of usage, testing whether it is a heavily quantized version. The results showed GPQA Diamond 92.9%, MMLU-Pro 88.7%, MATH-500 100% and AIME 2025 98.3%, in some cases above the published full-model range.
Highlights
- GPQA Diamond scored 92.9% and MMLU-Pro 88.7% on the discounted plan model
- MATH-500 reached 100% and AIME 2025 hit 98.3%
- Results sometimes exceed the published full-model range, with no sign of aggressive-quant damage
- OpenCode's dax responded that there is still work to do but that the offering deserves credit