Realtime AI News
Zhipu AI Releases GLM-5.3: Coding Nears Fable 5, Claims Strongest Open-Source Safety Model
Zhipu AI released its new open-source model GLM-5.3 today, saying its coding capability now sits much closer to Anthropic's Fable 5 and that testing surfaced a latent bug dormant for 40 years. The company also claims the model posts the strongest safety results among open-source models.
Zhipu AI released its new open-source model GLM-5.3 today, saying its coding capability now sits much closer to Anthropic's Fable 5 and that the model claims the title of strongest open-source safety model.
According to QbitAI's report, coding is where GLM-5.3 improves most, with the vendor describing the gains as "closer to Fable 5" — a framing widely read as a direct challenge to top closed-source models.
The release also featured a striking anecdote: the model reportedly surfaced a latent bug that had been dormant for 40 years during testing, showcasing its code comprehension and fault-locating ability.
On safety, Zhipu says GLM-5.3 posts the best evaluation results among open-source models, making "strongest open-source safety model" one of the launch's core selling points.
A hands-on review from ifanr published the same day reached a similar conclusion: in a week crowded with new model launches, GLM-5.3 fought its way back to the top tier of Chinese models and "pressed the reset button," suggesting the domestic competitive landscape may be reshuffled.
Beyond benchmark numbers, the significance is strategic. An open-source model closing the gap with closed-source flagships on both coding and safety gives developers and enterprise customers a more realistic alternative.
What to watch next: the open-source licensing and weights rollout for GLM-5.3, its coding performance in real engineering workloads, and whether it can keep matching Fable 5 in the next round of benchmarks.
Why it matters
GLM-5.3 narrows the gap between open-source and closed-source models on coding and safety, reshaping developer and enterprise model choices.
Nearby Updates
All08/14, 16:36
Baidu's Kuku AI Office Product Passes 25M MAU, Launches Standalone Client
Baidu Wenku and Baidu Netdisk's AI office product Kuku AI announced that its monthly active users have surpassed 25 million, alongside the launch of a new standalone office client. The move signals that Kuku AI is evolving from an embedded AI feature into an independent app carrying a complete AI office workflow.
08/14, 15:29
Compute demand up 10x in two years: robotics R&D devours GPUs on the road to the physical world
LimX Dynamics wrapped up its Chuangxue Camp 2026 hackathon with contestants running final simulation evaluations entirely on Alibaba Cloud's Wuying Linggou cloud workstations, as co-founder and CTO Chen Hua said R&D compute demand grew 5 to 10 times in two years. The platform moves GPUs and simulation environments to the cloud with template images and elastic compute, promising about half the cost and 80% faster deployment for robotics teams.
08/14, 05:13
Writer debuts new AI model and upgraded harness built on open-source GLM-5.2 to rein in token costs
Writer has introduced a new AI model built as a post-training variation on Z.ai's open-source GLM-5.2, alongside an upgraded harness meant to contain token costs. The company says the new system should provide deployment-ready capabilities at a much lower price.
08/14, 04:14
Databricks raises $5B at $190B valuation after investors bid up a planned $1B round
Databricks has closed a $5 billion round at a $190 billion valuation, CEO Ali Ghodsi told TechCrunch, after investors offered as much as $15 billion against the roughly $1 billion the company planned to raise. Ghodsi said Databricks accepted more than planned, blaming the scale on AI's high cost.