From the source
Together AI compares GLM-5.3 and Claude Fable 5 on DeepSWE, finding near-identical pass@1 accuracy but GLM-5.3 dominating cost and multi-attempt metrics.
Key findings: - Pass@1: Fable 5 69.7% vs GLM-5.3 69.0% (0.7 point gap, within noise) - Pass@2: GLM-5.3 81.1% vs Fable 5 77.1% - Pass@4: GLM-5.3 87.6% vs Fable 5 84.1% - Cost per rollout: GLM-5.3 $3.99 vs Fable 5 $21.63 (5.4x cheaper) - Per $100: GLM-5.3 solves 17 tasks, Fable solves 3 - Average runtime: GLM-5.3 35 min / rollout vs Fable 34 min (roughly even) - Per-task correlation: 0.65 (high agreement, making them near-substitutes) - GLM-5.3 leads on JavaScript (90 vs 75), Go (76 vs 71), TypeScript (61 vs 57), structured tasks (query&config, concurrency, program analysis) - Fable 5 leads on Rust (85 vs 70), Python (70 vs 66), data modeling, build&ops, protocol conformance - GLM-5.3 has higher coverage (87.6% vs 84.1%), Fable is slightly more reliable per attempt (82.0% vs 78.8%) - Both are safe with low regression rates (11% of failures regress existing tests) - Fable's premium edge is mainly Rust and serialization-heavy work; otherwise GLM-5.3 is recommended as the lower-cost substitute.
