GLM-5.3 Ties for #1 Open Model at Intelligence Index 60 — And Costs Less Per Task
By Eric Bush · August 20, 2026 · 6 min read
On August 14, 2026, Z.ai released GLM-5.3 and scored 60 on the Artificial Analysis Intelligence Index — tying Kimi K3 for the top open-model spot. The twist: it's the same base model as GLM-5.2. Every gain came from post-training. And it still costs $1.4/$4.4 per million tokens.
Most flagship launches ship a new architecture and a new pretrain. GLM-5.3 did neither. Z.ai took the existing 744-billion-parameter GLM-5.2 base and extracted the entire capability jump through extended post-training — reinforcement learning, tool-use tuning, and a heavy focus on long-horizon software engineering and cybersecurity/vulnerability analysis. The result is a model that trades benchmark blows with far more expensive frontier systems.
What "Intelligence Index 60" Means for Coding
The Artificial Analysis Intelligence Index is a composite of reasoning, coding, and knowledge benchmarks. A score of 60 puts GLM-5.3 level with Kimi K3 and within striking distance of the closed frontier tier. For coding specifically, that means it can hold its own on multi-file bug fixes and agentic tasks that used to require a premium model. The 1M-token context window means it can ingest a full repository without chunking — a real advantage on large-codebase SWE-bench-style tasks.
The Cost Case
GLM-5.3's official Z.ai pricing is $1.40 per million input tokens and $4.40 per million output, with cached input at $0.26/M. That's identical to GLM-5.2 — Z.ai raised the capability without raising the price. Compare that to a frontier proprietary model at $5/$25 (Claude Opus 5) or higher.
Model a typical agentic coding task: 60K input tokens (repo context + instructions) and 8K output tokens (reasoning + patch).
- GLM-5.3: 60K × $1.40/M + 8K × $4.40/M = $0.084 + $0.035 = $0.119/task
- Claude Opus 5 ($5/$25): 60K × $5/M + 8K × $25/M = $0.30 + $0.20 = $0.50/task
At roughly 4x cheaper per task with a comparable intelligence score, GLM-5.3 changes the routing calculus for teams doing high-volume agentic work. Across 500 tasks/day, that's $60 vs $250 — $69,000/year difference on one bucket of work.
The Catch: Weights Held Back
GLM-5.3 is billed as open-weight, but the weights were held back at launch — for roughly two weeks, access was through Z.ai's API and the GLM Coding Plan ($18–$168/month) only. If your plan depends on self-hosting to drive per-token cost to zero, verify the weights have actually shipped before committing. Until they do, you're on API pricing like everyone else.
This "post-train-only" release pattern is worth watching. If a lab can add a full tier of capability to an existing base and hold the price flat, it compresses the value of a brand-new pretrain — and it lets open-weight models close the gap to the frontier faster and cheaper than a from-scratch model ever could.
Should You Switch?
Route to GLM-5.3 for high-volume agentic coding, large-context repo work, and anywhere a 4x cost cut matters more than the last few points of accuracy. Keep a frontier model for the hardest novel problems and production-critical patches where a wrong answer is expensive. A tiered setup captures most of the savings while protecting quality where it counts.
Compare GLM-5.3 against your current model on real project sizes in our AI Cost Calculator.
Want to calculate exact costs for your project?
Frequently Asked Questions
What did GLM-5.3 score on the Artificial Analysis Intelligence Index?
GLM-5.3 scored 60, tying Kimi K3 for the top open-model position. Notably, it uses the same 744B base model as GLM-5.2, with all gains coming from extended post-training.
How much does GLM-5.3 cost?
Z.ai's official pricing is $1.40 per million input tokens and $4.40 per million output tokens, with cached input at $0.26/M — identical to GLM-5.2 despite the capability increase.
Is GLM-5.3 open-weight?
It's billed as open-weight, but the weights were held back for roughly two weeks after the August 14, 2026 launch. Initial access was via Z.ai's API and the GLM Coding Plan only.
Is GLM-5.3 cheaper than Claude Opus 5 for coding?
Yes, roughly 4x cheaper per task. A typical 60K-input/8K-output agentic task costs about $0.119 on GLM-5.3 vs $0.50 on Claude Opus 5 at $5/$25 rates.
Related Articles
Qwen3.8 Open-Source 2.4T Model: Second Only to Fable 5 at Zero API Cost
Alibaba's Qwen3.8 with 2.4 trillion parameters rivals Fable 5 at zero API cost. Compare open-weight self-hosting vs premium API pricing for AI coding.
Kimi K3 Released: 2.8T Open Source Model with 1M Context — What Coding Teams Pay to Run It
Moonshot's Kimi K3 is a 2.8 trillion parameter open-source model with native vision and a 1M token context window. We break down the real self-hosting hardware costs, when it beats API pricing, and how it compares to today's cheapest coding APIs.
China May Restrict AI Model Exports: How Open-Source Supply Shock Could Raise Global Coding Costs
Reuters reports China is planning to limit access to frontier AI models including open-weight releases. We analyze how restrictions on DeepSeek and Qwen could impact AI coding costs globally.