GLM-5.3 vs Kimi K3
Two Chinese-lab agentic models compared across weights, context, pricing, and coding benchmarks.
Updated August 28, 2026 · Independent comparison — GLM-5.3, Kimi K3 are separate products.
GLM-5.3
Z.ai's agentic coding & cybersecurity model
Z.ai's GLM-5.3 is an agentic coding and cybersecurity model built on the GLM-5 base, upgraded entirely through larger-scale post-training rather than a new pretrain. On Z.ai's own tests it reaches 28.3 on Terminal-Bench 3.0 (state-of-the-art among open-source models), 66.9 on DeepSWE v1.1, and 28.5 on Agents' Last Exam. It serves 128K output with always-on reasoning and function calling, and is priced at $1.40/$4.40 per million tokens ($0.26 cache read) through Z.ai's API (which speaks the OpenAI and Anthropic protocols) or the flat-rate GLM Coding Plan. Its 744B-total / 40B-active weights are scheduled to publish on Hugging Face on 2026-08-28, with the GLM-5.3-Flash variant already public; Z.ai has not published a raw context-window figure, though its Coding-Plan [1m] endpoint exposes a 1M-token window.
Kimi K3
Moonshot's 2.8T multimodal agentic model
Kimi K3 is Moonshot AI's 2.8-trillion-parameter model, released July 16, 2026 and billed as an "open 3T-class" system — though as of late August the weights have only been announced, with no release date, so it is API-first for now. It's a sparse Mixture-of-Experts (16 of 896 experts active) with native text, image, and video input, a 1M-token context, and a ~131K default output. Moonshot claims competitiveness with Anthropic's Fable 5, but publishes no independently verified benchmark table. It's served at $3.00/$15.00 per million tokens ($0.30 cache read) via Moonshot's API (model kimi-k3) and Kimi memberships, with a K3 Swarm Max variant for parallel multi-agent execution.
Bottom line
Two Chinese-lab agentic models compared across weights, context, pricing, and coding benchmarks. Choose GLM-5.3 if you want a coding/agentic model with published open-source SOTA on Terminal-Bench 3.0; choose Kimi K3 if you need native multimodal input — text, image, and video — in one model.
Model & Weights
| Feature | GLM-5.3 | Kimi K3 |
|---|---|---|
| Developer | Z.ai (Zhipu) | Moonshot AI |
| Parameters | 744B total / 40B active (MoE) | 2.8T total (sparse MoE, 16 of 896 experts) |
| Open weights | HF release scheduled 2026-08-28 (Flash variant already public) | Announced, no release date — API-first for now |
| Multimodal input | Not stated (text) | Text, image, video |
| Positioning | Coding + agentic + cybersecurity | Long-horizon coding, reasoning, agents |
Context & Output
| Feature | GLM-5.3 | Kimi K3 |
|---|---|---|
| Context window | Not published (1M via Coding-Plan [1m] endpoint) | 1M tokens |
| Max output | 128K tokens | ~131K default (configurable) |
| Reasoning | Always on (low / high / max) | Reasoning + agent workflows |
| Function calling | ✓ | Yes (OpenAI-compatible) |
Pricing (per 1M tokens)
| Feature | GLM-5.3 | Kimi K3 |
|---|---|---|
| Input | $1.40 | $3.00 |
| Output | $4.40 | $15.00 |
| Cache read | $0.26 | $0.30 |
| Subscription plan | GLM Coding Plan (Lite / Pro / Max) | Kimi membership (Moderato+; full 1M on Allegro+) |
Coding & Agentic Benchmarks
| Feature | GLM-5.3 | Kimi K3 |
|---|---|---|
| Terminal-Bench 3.0 | 28.3 (open-source SOTA) | Not published |
| DeepSWE v1.1 | 66.9 | Not published |
| Agents' Last Exam | 28.5 | Not published |
| Specific SWE-bench scores | Not published | Self-reported, not independently verified |
| Vendor positioning | SOTA open-source on Terminal-Bench 3.0 | Competitive with Fable 5 (Moonshot claim) |
Access & Ecosystem
| Feature | GLM-5.3 | Kimi K3 |
|---|---|---|
| Direct API | Z.ai (OpenAI + Anthropic protocols) | Moonshot API (kimi-k3) |
| Also on | GLM Coding Plan, ClinePass, OpenCode Go | ClinePass, OpenCode Go & Zen, Chutes TEE |
| Self-host weights | At/after 2026-08-28 release | Not yet (weights unreleased) |
| Multi-agent variant | Not stated | K3 Swarm Max |
| Released | August 2026 | July 16, 2026 |
The Verdict
Choose GLM-5.3 if...
- ✓You want a coding/agentic model with published open-source SOTA on Terminal-Bench 3.0.
- ✓You want lower per-token pricing ($1.40/$4.40) or the flat-rate GLM Coding Plan.
- ✓You want native Anthropic-protocol endpoints to drop straight into Claude Code.
- ✓You need long-horizon coding plus a cybersecurity-analysis focus.
- ✓You want downloadable weights soon (a 744B/40B-active MoE, publishing 2026-08-28).
Choose Kimi K3 if...
- ✓You need native multimodal input — text, image, and video — in one model.
- ✓You want the larger 2.8T-parameter model and Moonshot's Fable-5-class positioning.
- ✓You run parallel multi-agent workflows (K3 Swarm Max).
- ✓You're fine with an API-first model and don't need downloadable weights today.
- ✓You want a steep cache-read discount ($0.30) for repeated-context workloads.
Frequently Asked Questions
What is the difference between GLM-5.3 and Kimi K3?
Z.ai's GLM-5.3 is an agentic coding and cybersecurity model built on the GLM-5 base, upgraded entirely through larger-scale post-training rather than a new pretrain. Kimi K3 is Moonshot AI's 2.8-trillion-parameter model, released July 16, 2026 and billed as an "open 3T-class" system — though as of late August the weights have only been announced, with no release date, so it is API-first for now.
How much do GLM-5.3 and Kimi K3 cost?
GLM-5.3: $1.40. Kimi K3: $3.00. See the pricing table above for full plan details.
Should I choose GLM-5.3 or Kimi K3?
Choose GLM-5.3 if you want a coding/agentic model with published open-source SOTA on Terminal-Bench 3.0. Choose Kimi K3 if you need native multimodal input — text, image, and video — in one model.
1DevToolPoint either model at your whole workflow
GLM-5.3 and Kimi K3 both speak OpenAI-compatible APIs (GLM adds Anthropic-protocol endpoints), so you can wire either into an agent instead of a chat box. 1DevTool runs Claude Code, Codex CLI, Gemini CLI, Cline, and OpenCode side by side in persistent terminals — plug in a GLM Coding Plan or Moonshot key, swap models per task, and use the built-in HTTP client, 13-engine database client, and embedded browser. One-time $29, no subscription.