The lightweight, fast variant of Zhipu's GLM series, focused on low-cost high-concurrency inference.
Artificial Analysis released Endpoint Accuracy Index v1.1, tightening confidence intervals and reducing grader error on its long-context AA-LCR-25 benchmark while adding coverage for GLM-5.3 Flash and Muse Spark Open. Of 103 serverless endpoints benchmarked, 24 score statistically below the self-hosted reference, quantifying accuracy trade-offs from quantization, custom kernels and inference-stack tuning.
Meta Superintelligence Lab released Muse Ember, its lightest and most cost-efficient model yet, built for the repetitive work agents do all day: tool calls, coding subtasks, routing, extraction, and long document parsing. At 10 cents per million input tokens, Meta positions Ember as cheaper and smarter than Anthropic's Haiku and OpenAI's Luna on Artificial Analysis's composite benchmark, while staying price-competitive with open-source rivals DeepSeek V4.1 Flash and GLM 5.3 Flash. Ember is designed to pair with a larger model such as Muse Wonder inside an agentic harness.