GLMGolem Network Token प्राइस

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 the official API path of Zhihu. He then conducted a comparative experiment. Multiple GLM models hosted by Zhihu returned the same error message: 1214 Incorrect role information. When switching to DeepInfra hosting the same GLM weights, the error format changed. This indicates that it's not just that the model 'looks like GLM'; even the backend aligns with Zhihu. Previous model fingerprints also pointed to GLM. In 25 sets of text tests, the token count of Ox Alpha consistently differed by 75 from GLM-5.3; the visual token consumption from four different videos matched exactly with GLM-5V-Turbo. It can now be highly confidently asserted that the server for Ox Alpha comes from Zhihu, and the model itself is highly suspected to be part of the GLM series. However, it remains unconfirmed whether it is a new version of GLM-5.3 or an unreleased new model.

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.

According to monitoring by Dongcha Beating, Zhipu has officially launched the GLM-5.3 API. The API was not immediately available upon the model's release. Meanwhile, an independent evaluation by Artificial Analysis has been released: the Intelligence Index has surged from 53 for GLM-5.2 to 60, matching Kimi K3 and just 1 point lower than GPT-5.6 Sol. Once the equal-weighted scores are publicly disclosed as planned, it will rank alongside Kimi K3 as the highest-rated open-weight model in AA scoring. The most significant increase was seen in Agent, with the Elo of GDPval-AA v2 jumping from 1524 to 1770, a rise of 246 points, ranking second among all models, only behind Claude Opus 5's 1855, and surpassing Kimi K3's 1668 by over 100 points. The API pricing has not increased; input remains at $1.40 per million tokens, output at $4.40, and cached input at $0.26, the same as GLM-5.2. However, GLM-5.3 clearly utilizes tokens more effectively. AA measured that it outputs an average of about 18,700 tokens per task, which is 20% more than 5.2. Therefore, the actual cost per single task has risen from $0.44 to $0.68, an increase of about 55%. Even so, it remains cheaper than Kimi K3's $0.84 and GPT-5.6 Sol's $1.23 among comparable models.