GLMGolem Network Token प्राइस

According to Beating AI news, Terminal-Bench has released version 4.0, recalibrating the time, CPU, and memory used by agents when executing tasks. It has also fixed 19 tasks and removed 8 tasks that had saturation, refusal, public solution, or quality issues. The maximum execution time for all tasks has been standardized to 8 hours to reduce the impact of timeouts and environmental issues on scores. In the latest rankings, Opus 5 + Claude Code ranks first with 51.8%, followed by Fable 5 with 44.5%. GLM-5.3 + Claude Code achieved 41.8%, climbing to third place, surpassing GPT-5.6 Sol + Codex, which scored 37.3%. Among the top three, GLM-5.3 is the only model not from Anthropic. In Terminal-Bench 3.0, GLM-5.3 ranked fourth with 32.4%, trailing GPT-5.6 Sol's 34.6%; in version 4.0, GLM-5.3 has risen to third place, leading Sol by 4.5 percentage points.

According to Beating AI news, Zhipu has officially open-sourced its flagship model GLM-5.3 with 744 billion parameters, with FP8 and BF16 weights now available on Hugging Face. Unlike the previous generation GLM-5.2, this time it has replaced the MIT license with a new 'GLM-5.3 License'. The new license remains relatively permissive for ordinary developers, allowing them to use, modify, distribute, deploy, fine-tune, and even sell the model and its derivatives. However, Zhipu has added a restriction specifically for large 'model-as-a-service' companies: if the company and its affiliates have total revenue exceeding $10 billion for any consecutive 12 months, they must pass Zhipu's security review before commercially using GLM-5.3. Ordinary products embedding the model into specific functionalities are not subject to this restriction. GLM-5.3 was released two weeks ago, but Zhipu decided to conduct additional security assessments and enhancements due to the model's cybersecurity capabilities exceeding expectations before opening the weights. The model now supports frameworks such as SGLang, vLLM, Transformers, and KTransformers for self-deployment.

According to Beating AI news, Zhipu has officially open-sourced its flagship model GLM-5.3 with 744 billion parameters, with FP8 and BF16 weights now available on Hugging Face. Unlike its predecessor GLM-5.2, this version does not continue to use the MIT license but has introduced a new 'GLM-5.3 License'. The new license remains relatively permissive for ordinary developers, allowing them to use, modify, distribute, deploy, fine-tune, and even sell the model and its derivatives. However, Zhipu has added a restriction specifically for large 'model-as-a-service' companies: if a company and its affiliates have total revenues exceeding $10 billion in any consecutive 12-month period, they must pass Zhipu's security review before commercially using GLM-5.3. Ordinary products embedding the model into specific functions are not subject to this restriction. GLM-5.3 was released two weeks ago, but Zhipu decided to conduct additional security assessments and enhancements due to the model's unexpected growth in cybersecurity capabilities before releasing the weights. The model now also supports frameworks such as SGLang, vLLM, Transformers, and KTransformers for self-deployment.