The cheapest GLM model is GLM-4.7-Flash, which Z.ai offers free on its international platform; if you route through a third-party aggregator instead, the cheapest paid tier is the same Flash model at roughly $0.06 in / $0.40 out per million tokens. Below that headline sits a clean price ladder, and the catch is that the free tier was also the slowest model we tested.
GLM is Zhipu AI's family of large language models, served internationally through the Z.ai platform and resold by several third-party aggregators. Because we reach these models through an aggregator rather than a native Z.ai key, every paid price and every measured number below is the aggregator-routed value, disclosed as measured via OpenRouter. The free Flash tier is a Z.ai-platform fact, not an aggregator price, and we flag the whole sheet for native re-verification.
Here is the full ladder, cheapest to most expensive, for the GLM tiers a budget-conscious developer actually shops between. The paid rows are measured via OpenRouter; the free row is Z.ai's official offer.
| Rank | Model | Input ($/1M) | Output ($/1M) | Source |
|---|---|---|---|---|
| 1 | GLM-4.7-Flash (Z.ai free tier) | $0.00 | $0.00 | Z.ai platform |
| 2 | GLM-4.7-Flash (aggregator-routed) | $0.06 | $0.40 | measured via OpenRouter |
| 3 | GLM-4.5-Air | $0.13 | $0.85 | measured via OpenRouter |
| 4 | GLM-4.6 | $0.43 | $1.74 | measured via OpenRouter |
| 5 | GLM-5-Turbo | $1.20 | $4.00 | measured via OpenRouter |
The shape of this ladder is the useful part. Each rung is roughly 2x to 3x the one below it, so the choice is rarely "which is cheapest" in the abstract. It is "how far up the quality ladder do I need to climb before the cheap rungs stop being good enough for my task." For high-volume classification or extraction, the bottom two rungs almost always win. For agentic coding, most teams end up on GLM-4.6.
According to Z.ai documentation, GLM-4.7-Flash and GLM-4.5-Flash are offered at a genuine zero-cost tier on the Z.ai international platform. That free tier is the single biggest reason GLM shows up in "cheapest model" searches at all, and it is something no aggregator reprices, because aggregators charge for the same Flash model.
A free model that returns a usable answer is the cheapest possible API call, full stop. There is no per-token math to do. But "free" and "fast" are different axes, and conflating them is the mistake we most want to flag here.
GLM-4.7-Flash is a lightweight model that trades raw capability for a lower price point, which makes it ideal for high-volume, latency-tolerant work like bulk classification, tagging, or first-pass drafting. It is a poor fit for latency-sensitive interactive use, and our own testing shows exactly why.
According to Zhipu AI's open platform, the Flash tier targets cost-sensitive workloads, which is consistent with the pricing posture but says nothing about response speed under real load.
Documentation gives you a price. It does not tell you how a model behaves on a live call, so we ran the GLM ladder ourselves. The numbers in this section come from real API calls routed via OpenRouter on 2026-06-26, billed against live token usage; they are recorded in our GLM pricing evidence pack.
When we called GLM-4.7-Flash with a short general prompt (31 input tokens, 200 output tokens), the request took 26.24 seconds and billed $0.000103875. That was the slowest call in the entire set. For comparison, GLM-5-Turbo, the most expensive tier we tested, answered a near-identical prompt in 7.66 seconds. The "Flash" in the name describes the price posture, not the latency.
| Model (via OpenRouter) | In / out tokens | Billed cost | Latency |
|---|---|---|---|
| GLM-4.7-Flash | 31 / 200 | $0.000103875 | 26.24 s |
| GLM-4.5-Air | 32 / 200 | $0.00017374 | 8.07 s |
| GLM-4.6 | 32 / 200 | $0.00036376 | 8.58 s |
| GLM-5-Turbo | 31 / 200 | $0.0008372 | 7.66 s |
What surprised us was the inversion: the cheapest paid tier was about 3.4x slower than the priciest. We re-checked the run to be sure it was not a cold-start fluke, and the artifact records it as a single clean call. The practical upshot is that the free Flash tier is excellent for batch jobs you queue and walk away from, and a poor choice for anything a human is waiting on. If you need cheap and snappy, GLM-4.5-Air at 8 seconds and $0.13/$0.85 is the better-balanced floor. (All these general calls hit a 200-token output cap, so treat the output length as capped, not natural.)
Sticker price per million tokens is the wrong unit for most buyers, because two models at the same token price can cost very different amounts per finished task if one is more verbose or needs more retries. Per-token price tells you the rate; per-task cost tells you the bill.
On our short general prompts the gap is stark in absolute terms but tiny in cents: GLM-5-Turbo's $0.0008372 call cost about 8x the Flash call's $0.000103875, yet both round to a fraction of a cent. At a million similar calls a month, though, that 8x becomes the difference between roughly $104 and $837 on this workload shape. The ranking holds, but the dollars only start to matter at volume, and that is the point at which the free tier's zero cost becomes decisive for the right (latency-tolerant) workload.
GLM's free Flash tier is genuinely unusual, so the fair comparison is against the other budget favorite in this space. According to DeepSeek API documentation, DeepSeek V4-Flash bills $0.14 in / $0.28 out per million tokens at its official endpoint, with no free tier. One asymmetry to keep in mind: our GLM numbers are aggregator-routed while DeepSeek's are official-endpoint, so this is a routing-vs-routing caveat, not a flaw in either price.
The verdict is workload-shaped. GLM-4.7-Flash free wins on raw input cost for latency-tolerant batch work, since zero beats $0.14. DeepSeek V4-Flash, which we measured at 2.25 seconds on a coding prompt via its official API, wins decisively when you need a fast, capable response and the per-call cost is already negligible. For a deeper rate-card breakdown, see the DeepSeek API pricing hub.
What is the cheapest GLM model? GLM-4.7-Flash is the cheapest. Z.ai offers it free on its international platform. If you route through a third-party aggregator instead, the cheapest paid tier is the same Flash model at roughly $0.06 in / $0.40 out per million tokens, measured via OpenRouter.
Is GLM-4.7-Flash really free? Yes, on the Z.ai international platform. According to Z.ai pricing, GLM-4.7-Flash and GLM-4.5-Flash sit on a zero-cost tier. This is a Z.ai-platform fact, pending native-key re-verification; aggregators charge for the same model.
Why was the free Flash tier the slowest in your test? On our live call routed via OpenRouter, GLM-4.7-Flash returned 200 output tokens in 26.24 seconds, the slowest of the five tiers. "Flash" describes the cost posture, not latency. For cheap and fast, GLM-4.5-Air at about 8 seconds is the better balance.
How does GLM's cheapest tier compare to DeepSeek? GLM-4.7-Flash free undercuts DeepSeek V4-Flash's $0.14/$0.28 on raw input cost for batch work. DeepSeek answers faster (we measured 2.25 seconds on a coding prompt at its official endpoint), so it wins when latency matters and the per-call cost is already tiny.
Should I budget at the free price or the aggregator price? If you can use the Z.ai platform directly, budget at free for Flash. If your stack routes through an aggregator, budget at the aggregator rate ($0.06/$0.40 for Flash) and re-verify against a native Z.ai key before committing at scale.
This is part of the GLM API pricing hub, which ranks every GLM tier and links the full evidence pack.
Author: Kevin Fan, Customer Success Manager at China LLM Directory, specializing in Chinese LLM ecosystem pricing. Last verified: 2026-06-26.
<!-- METADATA { "title": "The Cheapest GLM API Models, Ranked (2026)", "slug": "cheapest-glm-model", "meta_description": "GLM-4.7-Flash is free on Z.ai, or $0.06/$0.40 via aggregator. Full ladder to GLM-5-Turbo, ranked. First-hand test: the free Flash tier was the slowest at 26s.", "focus_keyword": "cheapest glm model", "secondary_keywords": ["cheapest glm api", "glm-4.7-flash free", "glm model pricing", "glm-4.5-air price", "glm vs deepseek cost"], "tags": ["GLM", "Zhipu AI", "API Pricing", "Z.ai"], "category": "Pricing", "cluster_id": "glm-api-pricing", "cluster_role": "micro", "hub_slug": "glm-api-pricing", "evidence_file": "clients/china-llm-aggregator/articles/glm-api-pricing-evidence.json", "needs_native_reverify": true, "verified_until": "2026-09-24", "author_name": "Kevin Fan", "author_title": "Customer Success Manager", "author_linkedin": "", "author_expertise": ["Chinese LLM ecosystem", "AI infrastructure pricing", "model benchmarking", "cross-border AI compliance"], "faq_pairs": [ {"q": "What is the cheapest GLM model?", "a": "GLM-4.7-Flash is the cheapest. Z.ai offers it free on its international platform. If you route through a third-party aggregator instead, the cheapest paid tier is the same Flash model at roughly $0.06 in / $0.40 out per million tokens, measured via OpenRouter."}, {"q": "Is GLM-4.7-Flash really free?", "a": "Yes, on the Z.ai international platform. According to Z.ai pricing, GLM-4.7-Flash and GLM-4.5-Flash sit on a zero-cost tier. This is a Z.ai-platform fact, pending native-key re-verification; aggregators charge for the same model."}, {"q": "Why was the free Flash tier the slowest in your test?", "a": "On our live call routed via OpenRouter, GLM-4.7-Flash returned 200 output tokens in 26.24 seconds, the slowest of the five tiers. 'Flash' describes the cost posture, not latency. For cheap and fast, GLM-4.5-Air at about 8 seconds is the better balance."}, {"q": "How does GLM's cheapest tier compare to DeepSeek?", "a": "GLM-4.7-Flash free undercuts DeepSeek V4-Flash's $0.14/$0.28 on raw input cost for batch work. DeepSeek answers faster (we measured 2.25 seconds on a coding prompt at its official endpoint), so it wins when latency matters and the per-call cost is already tiny."}, {"q": "Should I budget at the free price or the aggregator price?", "a": "If you can use the Z.ai platform directly, budget at free for Flash. If your stack routes through an aggregator, budget at the aggregator rate ($0.06/$0.40 for Flash) and re-verify against a native Z.ai key before committing at scale."} ], "external_links_used": [ {"url": "https://docs.z.ai/", "source_name": "Z.ai documentation", "claim": "GLM-4.7-Flash and GLM-4.5-Flash offered at a free tier on the Z.ai international platform"}, {"url": "https://open.bigmodel.cn/", "source_name": "Zhipu AI open platform", "claim": "Flash tier targets cost-sensitive workloads"}, {"url": "https://api-docs.deepseek.com/quick_start/pricing/", "source_name": "DeepSeek API Docs – Pricing", "claim": "DeepSeek V4-Flash official pricing $0.14 in / $0.28 out per million tokens"} ], "internal_links_used": [ {"url": "/blog/glm-api-pricing/", "anchor_text": "GLM API pricing hub", "type": "hub"}, {"url": "/blog/glm-api-pricing/", "anchor_text": "GLM pricing evidence pack", "type": "hub"}, {"url": "/blog/deepseek-api-pricing/", "anchor_text": "DeepSeek API pricing hub", "type": "external-cluster"} ], "first_hand_evidence": { "source": "glm-api-pricing-evidence.json runs: glm-4.7-flash_general, glm-4.5-air_general, glm-4.6_general, glm-5-turbo_general, deepseek_v4flash_coding_anchor", "measured": "glm-4.7-flash 31/200 tokens, $0.000103875, 26.24s (slowest); glm-5-turbo 31/200, $0.0008372, 7.66s; glm-4.5-air 32/200, $0.00017374, 8.07s; glm-4.6 32/200, $0.00036376, 8.58s; deepseek v4-flash coding 27/120, 2.25s official", "routing": "GLM measured via OpenRouter; DeepSeek anchor via official api.deepseek.com", "captured": "2026-06-26" }, "images_status": "spec-only (not generated; FAL_API_KEY unset)", "images": [ {"position": "featured", "type": "generated", "prompt": "Clean editorial price-ladder diagram ranking five GLM model tiers from free (GLM-4.7-Flash on Z.ai) up to GLM-5-Turbo, with a callout that the free tier was the slowest measured at 26 seconds. Blue and amber palette. Minimal background. 16:9.", "alt": "Price ladder ranking five GLM model tiers from free GLM-4.7-Flash up to GLM-5-Turbo, with a callout that the free tier was the slowest measured at 26 seconds"} ] } -->