GLM 5.3 Artificial Analysis Score: 60 on the Intelligence Index — Explained
Quick answer: Artificial Analysis independently evaluated GLM 5.3 at 60 on its Intelligence Index (at max reasoning effort) — matching Kimi K3, three points behind the current leader Claude Opus 5 (63), and 8th of 181 models in its class (median: 35). The headline: GLM 5.3 reaches that score at the lowest cost per task of the top group — $0.68, versus $0.84 for Kimi K3 and $2.34 for Opus 5.
TL;DR
| Metric | GLM 5.3 |
|---|---|
| Intelligence Index | 60 (max reasoning effort) |
| vs Kimi K3 | Equal (60) |
| vs Claude Opus 5 (leader) | 3 points behind (63) |
| Class rank | 8th of 181 models |
| Class median | 35 |
| Cost per index task | $0.68 (lowest of top 3) |
| Parameters (AA record) | 753B |
| Evaluation cost | $1,238.50 on Z.ai API |
What the Score Means
On August 18, 2026, the independent evaluator Artificial Analysis published its first independent read on GLM 5.3. The model was tested at maximum reasoning effort — the setting Z.ai recommends for coding — and scored 60 on the Intelligence Index (v4.1.1).
The index aggregates nine evaluations: agentic real-world work tasks, agentic tool use, terminal coding, scientific reasoning and knowledge, graduate-level science questions, physics reasoning, knowledge reliability and hallucination, and long-context reasoning. GLM 5.3's 60 is the composite of that battery, run at a total evaluation cost of $1,238.50 on Z.ai's API.
Where It Lands
- Parity with Kimi K3 (60) — Moonshot's flagship, released July 16, 2026, remains the top-scoring open-weights model in the rankings; GLM 5.3 now shares that score (its license is proprietary for now).
- Three points behind Claude Opus 5 (63) — the current leader, released July 24, 2026. The gap shows what a fast post-training cycle did not close against a frontier that moved again in the spring.
- 8th of 181 models in its comparison class, well above the 35 median.
The Cost Angle: Cheapest Score-60
This is where GLM 5.3 separates from its score twins:
| Model | Index | Input/1M | Output/1M | Cost per task |
|---|---|---|---|---|
| GLM 5.3 | 60 | $1.40 | $4.40 | $0.68 |
| Kimi K3 | 60 | $3.00 | $15.00 | $0.84 |
| Claude Opus 5 | 63 | $5.00 | $25.00 | $2.34 |
GLM 5.3 reaches its score at the lowest cost per Intelligence Index task of the three — about 3.4× cheaper than Opus 5 per unit of measured intelligence.
The honest asterisk: GLM 5.3 is also the most verbose of the group, generating 170 million output tokens across the evaluation suite versus a 72 million median in its class. It's cheap per task and chatty — so your mileage varies by workload: long agent loops where token cost compounds will eat into the advantage; short-answer workloads won't.
Independent vs Vendor-Reported
The Artificial Analysis number carries different weight than Z.ai's launch table: the evaluator runs the suite itself rather than quoting vendor numbers. Z.ai's own benchmarks (Terminal-Bench 3.0 28.3, CyberGym 84.5, AutomationBench 48.2) are directional; AA's 60 is an independent composite across nine eval types. Both are worth reading — AA for the neutral score, Z.ai's table for the workload-by-workload splits.
What the Score Means for You
- If you're comparing GLM 5.3 vs Kimi K3: AA says equal intelligence, GLM 5.3 at roughly half the API price ($1.40/$4.40 vs $3.00/$15.00).
- If you're comparing vs Claude Opus 5: three index points behind, 3.4× cheaper per task — a classic value-vs-ceiling trade.
- If you're self-hosting: the 753B MoE base (per AA's record) needs serious hardware — plan multi-GPU or quantization.
FAQ
What did GLM 5.3 score on Artificial Analysis? 60 on the Intelligence Index (max reasoning effort) — matching Kimi K3, three points behind Claude Opus 5 (63), 8th of 181 models.
Is GLM 5.3 as smart as Kimi K3? Per Artificial Analysis, yes — both score 60. GLM 5.3 does it at roughly half the API price per token.
How much does GLM 5.3 cost on the API? $1.40 per 1M input and $4.40 per 1M output tokens (confirmed on Z.ai's API).
How many parameters does GLM 5.3 have? Artificial Analysis records 753 billion for the model; Z.ai hasn't officially confirmed an exact count.
Is GLM 5.3 better than Claude Opus 5? AA scores Opus 5 at 63 vs GLM 5.3's 60 — Opus 5 leads by three index points at 3.4× the cost per task.
Where do these numbers come from? Artificial Analysis Intelligence Index v4.1.1, published August 18, 2026.
Sources
- Unite.AI: GLM-5.3 Scores 60 on Artificial Analysis Intelligence Index, Matching Kimi K3 (August 18, 2026)
- Artificial Analysis: GLM-5.3
Last updated: August 18, 2026
