GLM-5.3 API Pricing: How Much Does the Open-Weight Model Cost?

What is GLM-5.3?

GLM-5.3 is Z.ai's frontier open-weight reasoning model, announced August 14, 2026. The launch post states it uses the same base model as GLM-5.2 and that every gain comes from post-training: more environments, more diverse tasks, and more compute spent on long-horizon training.

It is built for complex software engineering and long-horizon agent tasks, per the OpenRouter model description, with a 1M-token context window. Reasoning is always on and cannot be disabled; the API exposes three effort levels (low, high, max) with max as the default.

Z.ai is the company behind the GLM series, operating the Z.ai API platform. The launch post calls GLM-5.3 "the most capable open-weights model for coding" and says it achieves open-source SOTA on Terminal Bench 3.0 and Agents' Last Exam.

GLM-5.3 pricing: the verified rate card

The rates below were verified on the OpenRouter model page and the OpenRouter models API on August 25, 2026. GLM-5.3 is $1.40 per million input tokens and $4.40 per million output tokens, with cache reads billed at $0.26 per million.

OpenRouter lists a single provider for the model, Z.ai itself, so the listed price is the price you pay through that route. The model page reports a weighted-average input price of $0.30 per million after caching and discounts, with a 96.1% cache hit rate, which means most prompt tokens are billed at the cheap cache rate rather than the list rate.

For context, GLM-5.2 on the same marketplace is $1.19/M input and $3.74/M output, so GLM-5.3 carries a modest premium over its predecessor. Z.ai's own site does not publish a public pricing page; the OpenRouter listing is the accessible official rate.

What the benchmarks say

Z.ai's launch post includes a comparison table against Kimi K3, DeepSeek-V4 Pro-0813, Qwen3.8-Max, Claude Opus 4.8, Claude Fable 5, and GPT-5.6 Sol. On Terminal Bench 3.0, GLM-5.3 scores 28.3 versus GLM-5.2's 4.6, its largest coding jump, while Claude Fable 5 leads at 34.6 and Opus 4.8 at 33.7.

On DeepSWE v1.1, GLM-5.3 scores 66.9, ahead of DeepSeek-V4 Pro (62.7) but behind GPT-5.6 Sol (72.7) and Fable 5 (69.7). On the cyber benchmark CyberGym, GLM-5.3 tops the table at 84.5, ahead of Claude Mythos 5 (83.8) and GPT-5.6 Sol (83.6).

The ED-O-Meter run that hit Hacker News on August 23, 2026 (238 points) is narrower: a lap-style agentic benchmark where GLM-5.3 passed 100% of tasks at $0.28 per lap, about a fifth of GPT-5.5's cost, with a 16.3s time-to-first-token versus GPT-5.5's 13.2s.

GLM-5.3 vs GPT-5.6 Sol, Claude Sonnet 5, and DeepSeek V4: the price table

All prices are US dollars per 1 million tokens, verified August 25, 2026. OpenAI's GPT-5.6 Sol is on a promotional rate through at least November 21, 2026; DeepSeek V4 figures are the off-peak rates from the official docs, with peak-hour pricing double.

The takeaway for a cost-conscious buyer: GLM-5.3 at $1.40/$4.40 is about a third of GPT-5.6 Sol's promotional $4.00/$20.00 on input and less than a quarter on output, while DeepSeek V4 Flash remains the cheapest of the group at $0.22/$0.66 off-peak.

The honest catch: open-weight is still a promise

• Weights have not shipped yet. The launch post says weights release two weeks after launch, which points to around August 28, 2026. A HuggingFace search on August 25 found only community conversions (GGUF, MXFP4, abliterated variants), no official Z.ai release, so today the API is the only way to run GLM-5.3.
• Reasoning is always on. There is no non-thinking mode, so token spend on simple prompts is higher than on models with a thinking toggle.
• Single provider. OpenRouter routes every request to Z.ai directly; there is no multi-provider failover for this model yet.
• Benchmark framing. Z.ai's own table does not show GLM-5.3 ahead of Claude Fable 5 on most coding rows; the open-source SOTA claim is about open-weight models, not the closed frontier.
• Cyber capability is a compliance flag. The launch post highlights state-of-the-art vulnerability-discovery scores (CyberGym 84.5), which enterprise buyers may need to clear with security or compliance before adoption, a point the ED-O-Meter author repeats.

Who should use GLM-5.3

• Teams running long-horizon coding agents that want open-weight flexibility once the weights land, and a near-frontier model in the meantime via API.
• Cost-sensitive buyers migrating off GPT-5.6 Sol or Claude Sonnet 5: GLM-5.3's output rate is under half Sonnet 5's $10.00 and under a quarter of Sol's $20.00.
• Self-hosting teams willing to wait for the promised weight release around August 28, 2026, assuming the timeline holds.

Skip it if you need multi-provider redundancy, a thinking-off mode for latency-sensitive traffic, or a vendor with a published enterprise SLA. DeepSeek V4 Flash remains the cheaper workhorse for high-volume, low-risk jobs, and Claude Sonnet 5 the safer enterprise default.

At a glance

LLM API pricing per 1M tokens, verified August 25, 2026

FAQ

Is GLM-5.3 open source?

Not yet. Z.ai's launch post promises weight release two weeks after the August 14, 2026 launch, pointing to around August 28. As of August 25, 2026, only community conversions exist on HuggingFace, so the open-weight label is a commitment, not a shipped artifact.

How does GLM-5.3 pricing compare to GPT-5.6 Sol?

GLM-5.3 is $1.40/M input and $4.40/M output, versus GPT-5.6 Sol's promotional $4.00/M input and $20.00/M output (through at least November 21, 2026). That makes Sol about 2.9x more expensive on input and 4.5x on output.

Where can I buy GLM-5.3 API access?

The only current route is API access through OpenRouter (z-ai/glm-5.3, $1.40/$4.40 per 1M, live since August 19, 2026) or Z.ai's own API platform. Z.ai does not publish a public pricing page, so the OpenRouter listing is the accessible official rate.

Related reading

GPT-5.6 Sol pricing radar, Claude tool profile, ToolBistro radar index

Sources

• Z.ai GLM-5.3 launch post
• OpenRouter GLM 5.3 model page
• OpenRouter models API (pricing)
• ED-O-Meter benchmark (HN source)
• HN thread: GLM-5.3 for 1/5 the cost
• HuggingFace model search: glm-5.3