A bit more on how we fingerprinted OX Alpha:
• Compared its API surface against 418 OpenRouter models
• Full 7-part signature matched only OX Alpha, GLM-5.3 + its alias
• Exact same 10 supported parameters, including "top_k"
• Same 1,048,576 context / 131,072 max output
• Same "temperature: 1" + "top_p: 0.95"
• Same mandatory reasoning with low / high / max, default "max"
• GLM-5.2 does not share this configuration, it changed with 5.3 just days before OX appeared
• Then tested 50 adversarial tokeniser strings across Hebrew, Thai, Arabic, Korean, emoji, Unicode, digits, regex, whitespace etc.
• OX Alpha ↔ GLM-5.3: 50/50 exact token-count matches
• Gemini: 11/50
• DeepSeek: 9/50
• Kimi: 9/50
The important caveat: this does not prove OX is literally public GLM-5.3. OX has image + video input while public 5.3 is text-only.
So yes, it could be a newer/internal model, our finding is that it looks very strongly like a https://t.co/qxMwqW6bKJ model using the GLM-5.3-generation stack.
DeepSeek-V4-Flash-Vision-Exp is now live on the DeepSeek API Platform! 🚀
🔹 This experimental multimodal model matches DeepSeek-V4-Flash on text capabilities—including agents, reasoning, and world knowledge.
🔹 On multimodal agent benchmarks, V4-Flash-Vision-Exp makes a major leap over V4-Flash, bringing multimodal agent performance close to Opus-4.8.
Try it with model='deepseek-v4-flash-vision-exp'. DeepSeek Harness 0.1.1 was released today with out-of-the-box support for the new model.
1/n