Liquid AI's 2B d1 decision models beat same-size general models on classification benchmarks
JosephJacks_ · x · 2026-11-02
A developer benchmarked Liquid AI's d1 series 2B decision models against Qwen3.5-2B and InternVL3.5-2B on a single 5060 Ti (BF16, 4GB VRAM):
- Routing scanned docs into 16 types: d1-2B 88.8% vs Qwen 55.0%, InternVL 56.2%
- Flagging violence: d1-Omni-2B 97.3% vs Qwen 83.6%, InternVL 87.7%
- Spotting defective parts: d1-2B 70.8% vs Qwen 64.6%, InternVL 52.1%
- Latency: 0.13s per question for d1, vs 0.26s (Qwen) and 0.64s (InternVL)
The author notes it's not apples-to-apples—d1 models are decision specialists while the others are general chat models—but claims nothing else this small handles image-based decisions. Takeaway: small decision-specialized architectures offer strong accuracy-speed tradeoffs for classification workloads.
More from Infra
- Rumor: Musk to renegotiate TeraFab terms with TSMC, dropping self-build insistence — SumitGup · 2026-11-02
- Crusoe CEO says neocloud debt fears overblown amid 'tidal wave' of compute demand — nathanbenaich · 2026-11-02
- Kapitsa hits 346 tok/s on a 35B MoE and runs a 180B model on 48GB Macs via SSD offloading — DotaMate · 2026-11-02
- Texas Mega-Project Pipeline Tops $500B, Led by $100B Stargate AI Hub — MatthewChang · 2026-11-02
- SemiAnalysis: 30.3GW of planned 2028 US data center capacity at risk of pause after midterms — Eliv_nurotic · 2026-11-02
- B300 GPU rental prices plunge from $7.9/hr to $4.6/hr in a month as providers beg for customers — SumitGup · 2026-11-02