GLM-5.3 Ties Kimi K3 for Top Open-Model Ranking, Undercuts Rivals on Price — But Open Weights Delayed
Z.ai's GLM-5.3 ties Kimi K3 for the top spot among open models on the Artificial Analysis Intelligence Index, driven by a major leap in agentic task performance. The company is delaying the open-weight release by about two weeks, citing the model's unusually strong vulnerability-detection capabilities.
GLM-5.3 Matches Kimi K3 at the Top of the Open-Model Field
Chinese AI startup Z.ai has released GLM-5.3, a model that ties Kimi K3 for the top spot among open models on the Artificial Analysis Intelligence Index with a score of 60. That's seven points ahead of its predecessor, GLM-5.2, marking one of the larger single-version jumps recorded for an open-weight model this year.
The biggest gains show up in agentic performance. On the GDPval-AA v2 benchmark, GLM-5.3's Elo score climbed from 1,524 to 1,770 — a 246-point increase. That places it second overall on the benchmark, trailing only Anthropic's Claude Opus 5, which scored 1,855.
Pricing Undercuts Kimi K3
Artificial Analysis estimates GLM-5.3 costs $0.68 per task, roughly 1.5 times the $0.44 cost of GLM-5.2. Despite that increase, GLM-5.3 remains 19 percent cheaper than Kimi K3, which Artificial Analysis prices at $0.84 per task. Note that this is a per-task cost estimate from Artificial Analysis's benchmark suite, not a standard per-token API rate — Z.ai has not published input/output pricing per million tokens for GLM-5.3.
According to Artificial Analysis, the combination of near-frontier intelligence scores and lower per-task cost puts GLM-5.3 in direct competition with Western frontier models on both capability and price, closing a gap that has persisted between Chinese open-weight releases and models like GPT and Claude.
Open Weights Delayed Over Security Concerns
GLM-5.3 is currently accessible only through Z.ai's API. The company says it is delaying the release of the model's open weights by roughly two weeks. According to Z.ai, the delay stems from GLM-5.3's unusually strong ability to detect security vulnerabilities — the company says it wants to strengthen internal controls and give select security partners restricted early access before making the weights broadly available.
Z.ai has not detailed what specific vulnerability-detection capabilities triggered the delay, nor has it named the security partners receiving early access. The company's stated rationale is unverified beyond its own claim.
What This Means
GLM-5.3's benchmark results suggest Chinese open-weight labs continue to close the gap with proprietary frontier models, particularly on agentic tasks where GLM-5.3 now trails only Claude Opus 5. The price advantage over Kimi K3 reinforces a pattern where open-weight competition is increasingly fought on cost-efficiency as much as raw capability.
The delayed weight release is the more unusual story. If Z.ai's stated concern about vulnerability-detection capability is accurate, it signals that some open-weight models are now capable enough at security analysis that vendors feel compelled to slow-walk public access — a reversal of the usual dynamic where speed to open release is treated as a competitive differentiator. Whether this becomes a recurring pattern for future high-capability open models, or a one-off precaution, will depend on what Z.ai's security partners find during the two-week review window.
Related Articles
AWS launches SageMaker JumpStart Model Leaderboard covering more than 70 models
Amazon Web Services introduced the Amazon SageMaker JumpStart Model Leaderboard on November 18, 2026. It gives SageMaker Studio users one view of benchmark results, response performance and estimated instance pricing across more than 70 models.
GPT-6.2 Astra and Sol cut answer length 34 to 40 percent but fall sharply on Admits Uncertainty
An Artificial Analysis evaluation finds OpenAI's GPT-6.2 Astra and Sol produce answers 34 to 40 percent shorter than their predecessors and lead with the answer more often. Both score lower on uncertainty and transparency, and trail Anthropic's Claude Fable 5.2 and Opus 5.5 on the Transparency and Organization indexes.
Databricks launches public AI Index: open-weight models reach 30.9% of enterprise tokens, up from 7.3% in July
Databricks has launched the AI Index, a public resource combining usage data from 10,000+ companies on Unity Gateway with weekly coding-agent evaluations. Open-weight models grew from 7.3% of enterprise token traffic in July to 30.9% in November, and cost per task varies far more than token prices suggest.
NVIDIA Vera Rubin NVL72 posts over 10x GB300 NVL72 throughput per kW on MLPerf Endpoints v1.0
Provisional MLPerf Endpoints v1.0 results show NVIDIA Vera Rubin NVL72 delivering over 10x the throughput per kilowatt of GB300 NVL72 at a shared minimum interactivity of 50 tokens per second per user. The Vera Rubin results are pending MLCommons verification; the GB300 results are verified.
Comments
Loading...