Company · AI lab
Chinese AI company behind the M3 model family, listed in Hong Kong since January 2026.
MiniMax launched M3.1 Flash on October 19, 2026. See our report.
MiniMax's open-weight reasoning model with a 1M-token context, using hybrid 'lightning' attention to cut the cost of long reasoning.
MiniMax's natively multimodal mixture-of-experts model with about 428B parameters (23B active), a 1M-token context and MiniMax Sparse Attention.
A reasoning-first model from MiniMax aimed at agentic work, with text, image and video input and a 1M-token context window.
The AI stories that mattered on October 20, 2026: a federal DeepSeek ban, MiniMax M3.1 Flash, Tenstorrent's raise, Meta's Prometheus and the OpenCode breach.
thelatentdaily.com Editorial Team1 min read
Opinion: price-per-token cuts such as MiniMax's M3.1 Flash grab headlines, but agent bills depend on steps, retries and re-read context. A worked example.
thelatentdaily.com Editorial Team4 min read
Three cheap, fast models launched within days. We compare published prices, context windows and vendor benchmark claims, and what the numbers do not show.
thelatentdaily.com Editorial Team5 min read
MiniMax's M3.1 Flash costs $0.15 per million input tokens and scores 42 on the Artificial Analysis index. Open weights are due October 23; benchmarks are self-reported.
thelatentdaily.com Editorial Team4 min read