MiniMax M3 launched on June 1, 2026, and within a week became a new comparison point in the open-weight AI conversation. The Shanghai-based lab’s model scored 59.0% on SWE-Bench Pro, edging past GPT-5.5’s 58.6% on the same coding benchmark, according to a third-party comparison published by ofox.ai. MiniMax is charging as little as $0.30 per million input tokens, a fraction of what OpenAI and Anthropic charge for their flagship models. The company opened access to M3 through its first-party API on June 1, 2026, alongside listings on hosting platforms SiliconFlow, GMI and Novita. Fireworks AI added day-zero support, and OpenRouter listed the model within days, giving developers outside China quick access without a direct MiniMax account. Unlike its predecessor, M3 is MiniMax’s first model in the M-series built for native multimodality, accepting text, image and video input, a jump from the text-only MiniMax-M2.7 that shipped in March 2026. The model pairs that with a context window of 1,048,576 tokens. The market is crowded with Chinese open-weight launches this year, from DeepSeek V4 to Kimi K3 to GLM-5.2. MiniMax M3 adds native multimodality bolted onto a 1-million-token context window, priced