GLM 5.3 API Cost

Z.ai announced GLM 5.3 in August 2026 with headline benchmark jumps — Terminal-Bench 3.0 from 4.6 to 28.3, DeepSWE v1.1 from 46.2 to 66.9, and a best-in-benchmark 84.5 on CyberGym. If you're planning to build on the new flagship, the number that matters most isn't a benchmark score — it's the API cost per million tokens.

Here's the honest status: GLM 5.3 API pricing is not published yet. The official model page says the API "will be available soon," and the pricing page has no GLM-5.3 row. What is published — the cost model, GLM-5.2 flagship rates, and token-efficiency numbers — is enough to budget with today. This article explains the model, gives worked examples, and walks through the levers that decide the bill.

TL;DR

• Official GLM 5.3 API pricing is unannounced as of August 2026; the model API is expected to go live shortly.
• The current flagship, GLM-5.2, costs $1.40 per 1M input tokens, $0.26 per 1M cached input tokens, and $4.40 per 1M output tokens on the official pricing page.
• GLM 5.3 is reasoning-only (low/high/max) with a 1M-token context and 128K max output — output tokens and cache hits, not raw input, drive most of the bill.
• Z.ai reports GLM 5.3 completes tasks with fewer output tokens than GLM 5.2 at higher success rates — lower cost per completed task at an identical per-token price.

What is GLM 5.3?

GLM 5.3 is Z.ai's new flagship LLM, built on the same base as GLM 5.2 with improvements driven by post-training. Per the official model page, it is text-only, supports a 1M-token context window and 128K maximum output, and always operates with reasoning enabled at one of three effort levels — low, high, or max. Z.ai positions it for complex software engineering, long-horizon agent tasks, and security work, reporting a ~50% coding-performance gain over GLM 5.2 on Z.ai Code Bench. The glm-5.3 model ID and endpoints are documented, but the API is not live yet.

Is GLM 5.3 API pricing published yet?

No. As of August 18, 2026, Z.ai has not published a per-token price for GLM 5.3. The official pricing page lists GLM-5.2 as the top flagship, with no GLM-5.3 row — any number elsewhere claiming to be official is not from Z.ai. That doesn't stop you planning: GLM API pricing follows a stable structure, and the closest reference is GLM-5.2, the model GLM 5.3 builds on.

How GLM API pricing is structured

GLM API costs are billed per 1M tokens in three buckets (official pricing page):

Three things sit on top: caching is automatic (repeated prompts billed at the discounted cached-input rate, reported via usage.prompt_tokens_details.cached_tokens), tools are billed per use (Web Search $0.01 per use), and free models exist (GLM-4.7-Flash, GLM-4.5-Flash) for low-stakes work.

Reference prices: the GLM family today

Official rates (USD per 1M tokens):

Every flagship since GLM-5 has held input at $1.00–$1.40 and output at $3.20–$4.40 per 1M tokens — the most reasonable planning range for GLM 5.3, but strictly an assumption until Z.ai publishes the number.

Estimating GLM 5.3 API cost before the price is out

Use the same formula Z.ai's billing uses:

cost = (new_input / 1M × input_price)
     + (cached_input / 1M × cached_input_price)
     + (output / 1M × output_price)
     + (tool uses × per-use price)

Worked examples, using GLM-5.2 flagship rates ($1.40 / $0.26 / $4.40) as the stand-in:

1. Light reasoning (low effort, 5K input, 500 output): ≈ $0.009 per call.
2. Long-horizon repo session with a 1M-token context: first call pays 1M ÷ 1M × $1.40 = $1.40 for the full input. When 800K tokens cache on follow-ups, that portion drops to 800K ÷ 1M × $0.26 = $0.21 instead of $1.12.
3. Worst-case deep reasoning (max effort, 128K output fully used): 128K ÷ 1M × $4.40 = $0.56 per single response — why max_tokens and effort are real cost controls.

At Max effort on Z.ai Code Bench, GLM 5.3 reaches 34.5% at roughly 75K output tokens per task versus GLM 5.2's 23.4% at 96K. If that holds, cost per completed task drops even at an unchanged per-token price.

Track GLM 5.3 pricing and chat free on glm5.app/glm-5-3.

What will actually move the GLM 5.3 bill

• Reasoning effort — more effort means more reasoning tokens, and output is the most expensive line. Use low unless the task needs depth.
• Context length — with a 1M-token window, uncached input at $1.40/1M is the biggest fixed cost on first calls.
• Cache hit rate — stable prompts turn expensive input into $0.26/1M.
• max_tokens — a 128K output ceiling means one runaway response can cost over $0.50 at flagship rates.
• Tool use — Web Search at $0.01 per call adds up in agent loops.

Keep the GLM 5.3 bill low: a practical checklist

1. Start with reasoning_effort: low; raise it only when the task fails.
2. Set a hard max_tokens per task type — don't inherit the 128K ceiling by default.
3. Keep system prompts stable and push long documents into them so caching hits; confirm via cached_tokens.
4. Prototype on a free Flash model and measure token usage before moving to the flagship.
5. Add stop conditions to agent loops, and re-check the official pricing page the day the API goes live.

GLM Coding Plan vs. pay-as-you-go API

GLM 5.3 is served two ways today. The GLM Coding Plan — Z.ai's subscription for coding agents like Claude Code, Cline, and OpenCode — already serves GLM 5.3 on a points-based quota, and off-peak calls (including all of weekends) consume only 50% of the standard points. The pay-as-you-go model API has no announced price yet. If you need a raw API, you're waiting on the same announcement.

FAQ

When will GLM 5.3 API pricing be announced? Z.ai has not published a date. The official model page says the API "will be available soon," so watch the official pricing page and Z.ai's blog.

What is the GLM 5.3 API price? Not published as of August 18, 2026. The closest official reference is GLM-5.2 at $1.40 input / $0.26 cached input / $4.40 output per 1M tokens. Don't treat third-party numbers as official.

Does GLM 5.3 cost more than GLM 5.2? Unknown until published. Z.ai reports GLM 5.3 uses fewer output tokens at higher success rates, so per-task cost could fall at the same per-token price.

How does context caching change the cost? Repeated input is auto-detected and billed at the discounted cached-input rate ($0.26 vs $1.40 for the current flagship) — significant with a 1M-token window.

Is there a free GLM API model? Yes — GLM-4.7-Flash and GLM-4.5-Flash are listed as free on the official pricing page.

Can I use GLM 5.3 before the API launches? Yes, two ways: the GLM Coding Plan already serves it, and you can chat with it free on glm5.app.

Sources

• Official GLM-5.3 model page (Z.ai docs) — Specs, effort levels, "API coming soon" status, and Code Bench token-efficiency figures.
• Official Z.ai pricing page — Per-1M-token rates, cached-input rates, tool costs, and free models.
• Z.ai context caching documentation — How caching works, cached_tokens reporting, and discounted billing.
• Z.ai release notes — GLM 5.x timeline, including GLM-5.2 on June 16, 2026.
• Z.ai blog announcement for GLM 5.3 — Official announcement and benchmark context.

GLM 5.3 pricing and availability are pre-announcement (August 18, 2026) and may change; verify current rates on the official Z.ai pricing page. Rates for other GLM models are official at the time of writing.

Track GLM 5.3 pricing and chat free on glm5.app/glm-5-3 — updated the moment Z.ai publishes the official price.

Related Articles

• GLM 5.2 API: Endpoints, Authentication, and Python Integration Guide - GLM 5
• GLM 5.2 Cost Optimization: 6 Strategies to Reduce API Spend - GLM 5
• GLM 5.2 Free API: How to Get Free Tokens and Try GLM-4-Plus - GLM 5