Zhipu releases open-source model GLM-5.1, raises prices by 10% against the trend

Mainland media reports that Generalist big model developer Zhipu AI (02513) has officially released its new-generation open-source model GLM-5.1. According to OpenRouter, alongside this release, Zhipu’s GLM has once again increased its prices by 10%.

After the price adjustment, GLM-5.1’s cached-hit Token price in the Coding scenario is approaching the level of Claude Sonnet 4.6 from U.S. AI company Anthropic.

The report says this is the first time a domestic big model has achieved price parity with overseas top-tier manufacturers in a core scenario. From a year ago, when the industry generally fought for market share by cutting fees by 90% or more, to today, when performance premiums anchor international benchmarks—these changes indicate that domestic models are gradually moving away from the path of relying purely on low-price competition. They have begun to formally enter a new stage of value-based pricing, benchmarking performance against international standards.

GLM-5.1 continues to maintain a leading position in programming capabilities. In the comprehensive average scores across the three major code evaluation benchmarks—SWE-bench Pro, Terminal-Bench, and NL2Repo—it achieved the global third place, the top among domestic models, and the first among open-source models. In the SWE-bench Pro benchmark test, which is closest to real software development, it also achieved the first time a domestic model surpassed Opus 4.6, setting a new record for the best global result.

In addition, it has also made breakthroughs in long-horizon tasks. GLM-5.1 is the only open-source model that reaches 8 hours of continuous operation, and it is among the few modes globally with this long-horizon capability other than Claude Opus 4.6. By overcoming the current limitation of big models being mainly minute-level interactive, GLM-5.1 can continue and work autonomously for up to 8 hours within a single task. During the process, it can autonomously break down the task, repeatedly try and test, fix problems, and ultimately deliver complete engineering-grade results.

The release of GLM-5.1 marks that domestic big models are shifting from value-for-money competition to benchmarking capabilities against the world. It also suggests that AI is accelerating from efficiency tools toward a new form of productivity with independent output capabilities.