aiexpert
Home / News / Brief
Research · Aug 22, 2026, 06:36 AM · 4 sources

GLM-5.3 vs Claude Fable 5 on DeepSWE: 69% accuracy at 5.4x lower cost; routing strategy wins

<cite index="43-2">On the DeepSWE v1.1 coding benchmark, Claude Fable 5 scores 69.7% pass@1 at $21.63 per task, while GLM-5.3 scores 69.0% at $3.99 per task—a cost difference of 5.4x.</cite> <cite index="43-2">Per $100 spent, GLM-5.3 solves 17 tasks while Fable solves 3, making GLM the more cost-efficient path for budget-constrained agentic coding.</cite> DeepSWE tests long-horizon software engineering across 113 original tasks in 5 languages, <cite index="43-3">building contamination-free benchmarks that reflect how frontier coding agents actually perform in real software engineering work.</cite>

<cite index="43-2">The models separate once retries are allowed: GLM-5.3 leads pass@2 at 81.1% versus Fable 5's 77.1%, and extends that lead at pass@4 (87.6% vs. 84.1%).</cite> This pattern favors teams using routing or cascading strategies: run GLM-5.3 first at low cost, escalate to Fable 5's superior single-shot accuracy only when GLM fails. <cite index="43-2">Per-task correlation between the two is 0.65, the highest agreement in the comparison set, so they solve largely the same problems through different paths.</cite>

For production deployments, the trade-off is clear: GLM-5.3 is the cost optimization play, delivering 69% accuracy at a fraction of Fable's price; Fable 5 is the quality ceiling, with marginally higher single-shot performance and stronger reasoning on complex systems work. At 10,000-task scale, the cost gap ($39,900 vs. $216,300) justifies mixed routing logic. Neither model is universally superior—the choice depends whether your workload can absorb retries or demands first-attempt precision.

Sources

Everything this brief rests on
  1. 01 Primary source together.ai
  2. 02 together.ai together.ai “GLM-5.3 and Claude Fable 5 finish within noise of each other on DeepSWE accuracy. Fable 5 posts 69.7% pass@1 and GLM-5.3 posts 69.0%, a 0.7 point gap”
  3. 03 together.ai together.ai “GLM-5.3 costs $3.99 per rollout vs $21.63 for Fable 5. Per $100 spent, GLM-5.3 solves 17 tasks and Fable solves 3”
  4. 04 together.ai together.ai “Claude Fable 5 leads pass@1 by 0.7 points at 69.7%, but GLM-5.3 dominates retries. It leads pass@2 (81.1% vs. 77.1%) and pass@4 (87.6% vs. 84.1%)”