October 2026 Summaries

MiniMax M3.1 Flash is now available through a serverless, OpenAI-compatible API as a model designed for coding, autonomous agents, and complex workflows. The native multimodal Mixture-of-Experts model has approximately 428 billion total parameters while activating 23 billion per token, using sparse attention, quantization-aware training, and DSpark speculative decoding to reduce latency, memory use, and serving costs. It supports a one-million-token context window, image inputs, function calling, structured outputs, prompt caching, and adjustable reasoning effort. MiniMax reports strong results on agentic, coding, and computer-use benchmarks, including leading scores on Automation-Bench and LOCA-Bench 256K, while the provider also tested it with interactive simulations and SVG-generation prompts. Pricing is listed at $0.10 per million input tokens, $0.005 for cached input tokens, and $0.50 per million output tokens, and developers can access it by changing the model name in compatible API code.