Astra vs Fable 5.1 on real ML tasks: rigorous vs readable, with a mojibake twist
returnity · reddit · 2026-09-06
A developer ran a side-by-side comparison of Astra and Fable 5.1 (both xhigh) on a real ML text-processing and model-training workflow.
TL;DR: Astra codes more agentically, Fable more coherently. Astra's final outcome was slightly better with noticeably stronger scientific rigor and reproducibility; both improved F1/Accuracy by 0.02-0.04 after human feedback, showing neither has mastered the pipeline.
Astra's strengths:
- Stricter evaluation protocol (70/15/15 split with held-out validation model selection vs Fable's 80/20 + test F1).
- Hit a gensim 4.4 compiled-kernel bug: Fable hid the stderr notices; Astra root-caused it and fixed the environment with a compatible gensim/NumPy/SciPy downgrade.
- Wrote hardened training code: isolated uv venv rebuild, SHA-256'd corpus, split manifest, run-summary., even headless-browser QA screenshots.
- Deployed subagents effectively; the notebook-reviewer caught a real tokenization bug and left a regression test.
Fable 5.1's comeback:
- Astra confidently shipped a verifiable encoding defect, decoding UTF-8 as Windows-1252 and producing mojibake in currency symbols; Fable verified UTF-8 and got it right.
- Fable's analysis report was more insightful, running an ablation that surfaced an expensive no-op preprocessing step.
- More idiomatic, readable code that better followed coding-conventions.md, and better scoping. The author now picks Fable 5.1 for prose—something they hadn't said about Claude in a while.
Related event: Astra vs Fable 5.1: rigorous agent vs human-like intuition(4 posts)→
More from Models
- Meta researcher: horrifying infra, not talent, is why OSS can't catch closed frontier — A_K_Nain · 2026-10-23
- TypeSafe's Jev: a "Decision Model" LLM That Returns Floats at $0.042/M Input Tokens — ariG23498 · 2026-10-23
- Microsoft's MAI-Base-2 scaling ladder pegs the L42 run at ~1.8×10^23 FLOPs — teortaxesTex · 2026-10-23
- Local model agentic coding benchmark: 27B matches 125B, tool-calling is the lever — Joralk_ · 2026-10-23
- NVIDIA open-sources PersonaPlex-7B, a full-duplex voice AI model that listens and talks simultaneously — solyarisoftware · 2026-10-23
- NVIDIA's KumoRFM-2: a foundation model that predicts churn, fraud and demand straight from relational databases — solyarisoftware · 2026-10-23