What GLM-5.3 costs for coding: API, cache, and plan

At a glance

Three workloads, priced

The session workload used across this blog is 100,000 tokens of fresh input, 400,000 written to the cache, 2 million read from it, and 50,000 of output. On GLM-5.3 that comes to $1.44: $0.14 of input, $0.56 for the tokens written to cache at the ordinary input price, $0.52 of cache reads, and $0.22 of output. Without caching, the same tokens would cost $3.72.

A large one-off review, 150,000 input tokens with no cache hits and 10,000 output, costs $0.25. An output-heavy generation, 30,000 in and 80,000 out, costs $0.39. On GLM-5.3-Flash the three workloads cost $0.16, $0.03, and $0.04, which is why the GLM-5.3 vs GLM-5.3-Flash comparison is mostly a question of quality.

How Z.ai's caching differs

Anthropic charges a premium to write the cache and then bills hits at a tenth of input. Z.ai does the opposite: writes cost ordinary input, and a hit on GLM-5.3 costs $0.26 per million, about 19% of the $1.40 input price. Caching therefore never costs more than not caching on GLM, but each read saves less than on Claude, so the share of a GLM session spent on cache reads is larger. The cache write premium entry explains the Claude side.

Caching is automatic, with no configuration, and Z.ai lists cached storage as free for a limited time without an end date. On a Coding Plan, Z.ai keeps reasoning from earlier turns by default, which raises cache hits.

Reasoning effort is a cost setting

GLM-5.3 always reasons, at low, high, or max, and max is the default. Z.ai's own coding evaluation cites about 75,000 output tokens per task at max against about 50,000 at high. At $4.40 per million output tokens, those extra 25,000 tokens add about $0.11 per task, so routine edits at high effort cost noticeably less. In Claude Code, /effort sets the level, as the Claude Code setup guide explains.

API or Coding Plan

A hundred and ten sessions a month, five a day over twenty-two working days, would cost $158.40 on GLM-5.3 at API rates. The GLM Coding Plan meters the same session at 805 credits at peak hours, so Lite's weekly allowance covers about twelve of them for $18 a month, Pro about seventy-four for $80. For steady use the plan is far cheaper per session; for occasional use, or to mix models, paying per token through Z.ai or OpenRouter avoids a monthly fee.

Sources

Checked October 3, 2026. Plans and limits change; the linked pages are the authority.

1. Z.ai docs: Pricing
2. Z.ai docs: Context caching
3. Z.ai docs: GLM-5.3
4. Z.ai docs: Thinking mode
5. Z.ai docs: GLM Coding Plan overview