Qwen 3.8 27B vs Flash Next 125B on a threat-intel platform: both fail, 27B wins
Character-Apple-8471 · reddit · 2026-10-15
The author gave both models the same prompt: build a real-time cyber threat intelligence platform (FastAPI, async ingestion, WebSockets, Three.js globe) with strict data-integrity requirements—real sources, preserved attribution, no invented attacks. Neither fulfilled the full prompt, but the author would rather repair the 27B version. Flash Next 125B delivered more features and tests (34 vs 16 passing) but broke core data contracts: its Feodo adapter parsed 0 of 5 records yet reported success; its DShield adapter read the subnet prefix (24) as the activity count (344), corrupting volume-based scoring; its Cloudflare Radar request/parser mismatched the documented API. Both mishandled ThreatFox and had WebSocket backpressure issues. Artifact comparison, not a controlled study.
More from Models
- NCM: new text generation paradigm up to 31x faster than diffusion LMs with better quality — simonguozirui · 2026-10-15
- Gemma 4 26B priced at $0.07/M input tokens, ~40% below comparable providers — testingcatalog · 2026-10-15
- Epoch AI: GPT-6 Astra tops frontier models in Slay the Spire 2 but still loses to top humans — sandersted · 2026-10-15
- Google execs cite early signs of recursive self-improvement behind 3-4 week Gemini cadence — danielrock · 2026-10-15
- Claude systematically debunks the speculative EDAR-selenium hypothesis about East Asians — teortaxesTex · 2026-10-15
- Meta's Muse Wonder (Max) lands at #12 in Agent Arena at 2.1x the cost of its predecessor — arena · 2026-10-15