SlopCodeBench Oct 2026: Astra leads at 16.3%
Original: SlopCodeBench 10/2026 – Opus 5.5, Astra wins overall, xhigh considered harmful
Why This Matters
Shows frontier coding models still fail over 83% of strict real-world code tasks.
GPT-6 Astra scored 16.3% strict pass rate on SlopCodeBench in October 2026, topping gpt-5.5's May score of 14.8%. Frontier models tested ranged 12.2–16.3%, with Opus 5.5 also featured.
HumanLayer's SlopCodeBench (SCB) October 2026 update puts GPT-6 Astra at the top with a 16.3% strict pass rate — a 1.5-point improvement over gpt-5.5, which had held the high mark at 14.8% since May. The benchmark's 'strict pass rate' measures how often a model clears every test case in a checkpoint without breaking regression tests — a notably unforgiving standard. All frontier models in this round landed in a tight 12.2–16.3% band. Models tested include Opus 5.5, Fable 5.1, SOL 6 and 6.1, and Astra. The report also flags 'xhigh' settings as counterproductive — a finding to be detailed in Part II, a deeper qualitative look at 'slop patterns' co-authored by Gabe Orlanski and the SCB team. Dex Horthy authored this Part I installment.