Particle-1 trails Opus 5.5 by 3.4 points on memory eval but costs 1/23 per correct answer
garrytan · x · 2026-11-02
Garry Tan published a gbrain-evals finding on LongMemEval (500 questions over long chat histories), where a shared memory system retrieved identical candidate conversations for every model. particle-1 scored 457/500 (91.4%) vs Opus 5.5's 474 (94.8%) and GPT-6.1 Sol's 464 (92.8%); the 3.4-point gap is statistically real (95% range 1.4–5.6).
The striking part is cost per correct answer at list price: Opus 5.5 $0.0993, GPT-6.1 Sol $0.0381, particle-1 just $0.0043 — $1 buys 10 right answers from Opus but 230 from particle-1, roughly 23x cheaper.
Caveats: one early test, public questions, a single AI grader, and all models clustered at 91–95%, leaving little headroom to differentiate.
More from Infra
- Blackwell servers now fetch $2.4M each in China, prices tripled in four months — teortaxesTex · 2026-11-02
- Apple's MLX and mlx-lm: Local LLM Inference on Apple Silicon — goyalshaliniuk · 2026-11-02
- llama.cpp: Running GGUF-Quantized LLMs on CPUs, GPUs and Apple Silicon — goyalshaliniuk · 2026-11-02
- 5 Inference Engines Powering Modern AI: vLLM, SGLang, TensorRT-LLM, llama.cpp, MLX — goyalshaliniuk · 2026-11-02
- Nvidia's 27-year market cap history: from $800M to $5T, with 70% crashes along the way — mitchdeg · 2026-11-02
- 38 hours on one RTX 3060: SLQ quantization edges out AtomicChat GGUF at same size — maximelabonne · 2026-11-02