Fable 5.2 solves over half of expert long-horizon engineering tasks, surprising evaluators
Jsevillamol · x · 2026-11-14
Carlos Bort shared an update noting Fable 5.2 performs surprisingly well at engineering: on long-horizon tasks designed by experts, the model solved more than half of the runs. The team plans to benchmark newer models and build harder evaluation tasks next.
More from coding & agent
- Dev picks paper as best BYO-agent MCP experience with tens of concurrent subagents — jakedahn · 2026-11-14
- gpucanvas open-sourced: thousands of live DOM nodes on a GPU-drawn canvas — Vjeux · 2026-11-14
- WebGPU canvas renders 7,000 live DOM frames with 2M+ elements at 120fps — Vjeux · 2026-11-14
- Free open-source "AI Engineering from Scratch" course teaches algorithms from raw math — ghumare64 · 2026-11-14
- Hugging Face Introduces Decision Models That Output Probability Distributions Without Generating Text — ariG23498 · 2026-11-14
- Inside the MI300X: 23-min visual guide from compute chiplets to a Gluon GEMM — saurabhtwq · 2026-11-14