be z. ai - come up with ox alpha - market it as a garage project - in reality test glm-5.3-flash in stealth - make everyone talk about it what happened: on aug 20 it they dropped on openrouter as stealth/ox-alpha it had no company no paper $0 in $0 out 1m context text + image + video in > early 10-task screenshots made it look like it cooked claude > full coding runs were closer to gpt-5.6 sol, not opus > their real edge was the pricing for 6 days it was the most used model on openrouter 20 trillion+ tokens then on aug 26 they unmasked it what it actually is: - glm-5.3-flash - 320b moe / 18b active - 1,048,576 token context - 131k max output - tool calling + json - mit weights on hugging face - reads video, does not generate video ------------------------> the free endpoint is dead stealth/ox-alpha does not route anymore use this instead: z-ai/glm-5.3-flash ------------------------> still super cheap vs claude current price (until sep 9 2026, utc+8): $0.075 / 1m input $0.25 / 1m output after that: $0.15 in $0.50 out getting started takes less than 2 minutes: 1. open https://t.co/1hNdUZ80vv 2. or use https://t.co/p9xSlXYWUA api / glm coding plan 3. sign up using gmail or wallet 3.1. swap any old stealth/ox-alpha calls to the new id 4. if you have a gpu, grab the mit weights and self host the best part? you can still run a 1m context multimodal coding model for cents while people are paying $20-50/mo for the same class of work