ALPHAAlpha Finance 価格

According to Beating AI news, Zhipu has confirmed that Ox Alpha is a new version of the GLM series, with model weights set to be released tonight. The community had previously pointed towards Zhipu based on API errors and tokenizer clues, and this confirmation solidifies that. Ox Alpha anonymously logged into OpenRouter last weekend, focusing on programming and long-duration agent tasks, while supporting text, image, and video inputs. After its launch, it quickly became the most used model on OpenRouter, with usage exceeding that of DeepSeek by more than twice. OpenRouter stated that this is the largest model release in the platform's history. The multimodal capability is also a significant change to note. Zhipu has previously kept the main GLM and GLM-V visual models separate, with models like GLM-5 supporting only text, while images and videos were handled by models like GLM-5V-Turbo. This time, Ox Alpha is directly recognized as a new GLM version that can handle images and videos. If the final official model does not carry the 'V', it indicates that Zhipu is beginning to merge the two model lines. Ox Alpha will remain free for another week, with pricing details yet to be announced.

According to Beating AI news, Zhipu has confirmed that Ox Alpha is a new version of the GLM series, with model weights to be publicly released tonight. The community had previously pointed to Zhipu based on clues from API errors and tokenizers, and this confirmation solidifies that. Ox Alpha anonymously logged into OpenRouter last weekend, focusing on programming and long-duration agent tasks, while supporting text, image, and video inputs. After its launch, it quickly became the most used model on OpenRouter, with usage exceeding that of DeepSeek by more than double. OpenRouter stated that this is the largest model release in the platform's history. The multimodal capability is also a noteworthy change. Zhipu has previously kept the main GLM and GLM-V visual models separate, with models like GLM-5 supporting only text, while images and videos were handled by models like GLM-5V-Turbo. This time, Ox Alpha has been confirmed as a new GLM version that can simultaneously process images and videos. If the final official model does not carry the 'V', it indicates that Zhipu is starting to merge the two model lines. Ox Alpha will remain free for another week, with details on future pricing yet to be announced.

According to Beating AI news, the identity of Ox Alpha is becoming increasingly difficult to conceal. Community developer Chetaslua sent a deliberately erroneous API request to the model via OpenCode. The backend unexpectedly output a Java stack trace, which included paas/v4/chat, directly aligning with Zhipu's official API path. He then conducted a comparative experiment. Multiple GLM models hosted by Zhipu returned the same 1214 Incorrect role information error. When switching to DeepInfra hosting the same GLM weights, the error format changed. This indicates that it is not just that the model 'looks like GLM'; even the backend aligns with Zhipu. Previous model fingerprints also pointed to GLM. In 25 sets of text tests, the token count of Ox Alpha consistently differed from GLM-5.3 by 75 tokens; the visual token consumption from four different videos matched perfectly with GLM-5V-Turbo. It can currently be highly confirmed that the server of Ox Alpha comes from Zhipu, and the model itself is highly suspected to be part of the GLM series. However, it remains unverified whether it is a new version of GLM-5.3 or an unreleased new model.