Z.ai’s GLM-5.3 has been evaluated by Synthetic Evaluation at 60 on its Intelligence Index, the unbiased evaluator reported on August 18, 2026, putting the Chinese language lab’s latest reasoning mannequin degree with Moonshot AI’s Kimi K3 and three factors behind Anthropic’s Claude Opus 5, the present chief at 63.
The rating covers GLM-5.3 operating at its most reasoning effort, the setting Z.ai recommends for coding work. At 60, it sits properly above the 35 median of the 181 fashions in its comparability class, and eighth within the class total.
The result’s the primary unbiased learn on a mannequin that Z.ai launched on August 14, 2026 with an uncommon building: GLM-5.3 makes use of the identical base mannequin as GLM-5.2, with each functionality acquire coming from post-training somewhat than a brand new pretraining run.
How GLM-5.3 Acquired Right here
Z.ai’s launch submit describes a month of scaling reinforcement studying on long-horizon activity environments: extra environments, extra numerous duties, extra compute, all on the coaching stack the lab constructed for GLM-5.2. The corporate reported that strategy moved Terminal-Bench 3.0 from 4.6 to twenty-eight.3 and DeepSWE v1.1 from 46.2 to 66.9, and its personal submit paperwork a diffusion of outcomes throughout coding, cybersecurity, and agentic benchmarks towards Kimi K3, Claude Opus 4.8, Claude Fable 5, and GPT-5.6 Sol. Unite.AI coated the launch and its emergent cybersecurity leads to element when the mannequin shipped final week.
The Synthetic Evaluation quantity carries totally different weight than these vendor-reported tables, as a result of the evaluator runs its suite itself. Its Intelligence Index v4.1.1 aggregates 9 evaluations spanning agentic real-world work duties, agentic instrument use, terminal coding, scientific reasoning and information, graduate-level science questions, physics reasoning, information reliability and hallucination, and long-context reasoning. GLM-5.3’s 60 is the composite of its run throughout that battery, at a complete analysis price of $1,238.50 on Z.ai’s API.
The place GLM-5.3 Lands on the Leaderboard
The parity with Kimi K3 is the headline comparability. Kimi K3, launched July 16, 2026, scores the identical 60 on the index and stays the top-scoring open-weights mannequin in Synthetic Evaluation’s rankings, a place GLM-5.3 now shares in rating if not in license. Moonshot opened Kimi K3’s weights below a revenue-tiered license in July 2026; GLM-5.3 is listed as proprietary for now, with Synthetic Evaluation recording 753 billion parameters for the mannequin.
Claude Opus 5, launched July 24, 2026, nonetheless leads the index at 63. The three-point hole between GLM-5.3 and the chief is the space a quick post-training cycle didn’t shut, towards a frontier that has itself moved for the reason that spring.
Worth is the place the 2 60-scorers diverge sharply. GLM-5.3 prices $1.40 per million enter tokens and $4.40 per million output tokens on Z.ai’s API, towards $3.00 and $15.00 for Kimi K3 on Moonshot’s. Per Intelligence Index activity, that works out to $0.68 for GLM-5.3 versus $0.84 for Kimi K3 and $2.34 for Claude Opus 5. GLM-5.3 reaches its rating on the lowest price per activity of the three, although additionally it is probably the most verbose of the group, producing 170 million output tokens throughout the analysis suite towards a 72 million median in its class.
What Ships Subsequent
GLM-5.3 is accessible by means of Z.ai’s API and has been rolled out to all GLM Coding Plan subscribers. The mannequin requires pondering to be enabled, with three effort ranges, and Z.ai warns that purposes nonetheless calling it with pondering disabled will fail till migrated.
The weights are the remaining piece. Z.ai has dedicated to releasing them two weeks after the August 14, 2026 launch, as soon as security analysis and hardening are full, which might put GLM-5.3 alongside Kimi K3 as an open-weights choice on the 60 mark on the index, at roughly half the per-token worth.
