iFlytek Launches Spark X2.6 Hybrid Edge-Cloud LLM Trained on Domestic Chips, Open-Sources 4B Model
智东西 · wechat · 2026-10-22
iFlytek unveiled Spark X2.6, an edge-cloud hybrid LLM trained entirely on domestic Chinese compute, comprising a cloud 293-A30B model (1M-token context) and an open-sourced on-device 4B model. Pricing: 0.9 RMB/M input tokens, 3.6 RMB/M output—claimed cheaper than DeepSeek-V4.1-Flash-1017.
Founder Liu Qingfeng argued full-stack self-controllability is essential for AI safety and cost efficiency, predicting 60% of tasks will run locally, 30% edge-cloud collaborative, 10% cloud-only. The 4B model reportedly beats Qwen3.5-4B/9B and Gemma4-E4B; the prior X2.5-4B hit 2.56M downloads. Training on the Feixing-2 platform saw 4x efficiency gains; KVCache hit rate rose from 63% to 95.7%.
Also announced: SparkAudio1.0 speech foundation model, AI recording ring (¥1,299), AI glasses (from ¥4,299), AIPC laptops (up to 160 TOPS), office all-in-one (2240 TOPS), WinWork digital employee platform, teacher assistant, and robot brain platform V3.0.
More from Embodied
- MIT CSAIL's X-Hinges 3D-Prints Everyday Objects With Built-In Sensing Across Three Motion Axes — MIT News AI · 2026-10-23
- Formic raises $50M Series B to scale physical AI across 100+ US factories — garrytan · 2026-10-23
- fal starts fine-tuning a physical robot to draw, results next week — OdinLovis · 2026-10-23
- PEARL demo: an LLM agent that checks experiment results, revises plans, and re-tests — KevinKaichuang · 2026-10-23
- CU Boulder's ATLAS Institute Opens Tenure-Track Robotics Professorship — ryosuzk · 2026-10-23
- MiTaS multi-resolution tactile imitation learning hits 80% success vs 35.5% vision-only baseline — Jan_R_Peters · 2026-10-23