What Is GLM-5? Zhipu's Open Agent Model (2026)

GLM-5 is Zhipu's open-release large language model tuned for programming and long-duration agent work, and on OpenRouter it prices at roughly $0.60 input and $1.92 output per million, the priciest open-weight model in our July 2026 test set. It is also the slowest one we clocked, which makes it an interesting case: a capable coding-and-agent model whose cost and latency profile sits well above its cheaper open-weight peers, so it pays to know exactly what you are buying.

What is GLM-5 (verified 2026-07)

GLM-5 is an open-weight large language model from Zhipu that targets programming and long-horizon agent tasks rather than general chat. The GLM line has always leaned toward tool use and structured reasoning, and this release continues that focus, which is why it keeps surfacing in agent-builder shortlists next to the cheaper Qwen and DeepSeek open weights.

According to Z.ai, the international home of Zhipu's GLM models, the GLM family is positioned around coding and agentic workloads, the kind of multi-step task where a model plans, calls tools, and works over a long context before returning a result. That positioning is the useful frame for GLM-5: it is not trying to be the cheapest token on the market, it is trying to be the one you trust to drive an agent loop end to end.

Two facts pin the model down for a buyer. GLM-5 is developed by Zhipu AI, the Beijing lab also known internationally as Z.ai, and GLM-5 is served on OpenRouter under the model string glm-5-20260211, which is the exact build we tested. Both are worth recording, because a model name without a served-build string is hard to budget or reproduce against later.

GLM-5 pricing vs other open-weight models

Here is where GLM-5 lands against the other newly-released open-weight models we priced on the same routing marketplace on the same day. Every figure is measured via OpenRouter and carries a native-reverify flag, since routed rates can differ from a model's first-party endpoint.

The output column is where GLM-5 stands apart, and not in the direction most buyers hope. At $1.92 per million output, GLM-5 is roughly 19 times more expensive than Qwen3-235B-A22B and more than double MiniMax-M2.5, which is the part most cost tables gloss over: among the "cheap Chinese open-weight" cohort, GLM-5 is the premium tier. According to DeepSeek's official API pricing, the V4-Flash endpoint lists at $0.14 input and $0.28 output per million, so as a first-party comparator GLM-5's routed output rate sits nearly 7 times above even that aggressively-priced number.

One model in this space we could not price directly is Moonshot's Kimi-Dev-72B, an open coding model that was not callable on the routers we tested. We flag it only as sourced context: Moonshot's own materials report it at roughly 60.4% on SWE-bench Verified. We did not run it and are not presenting that number as first-hand, so treat it as a vendor claim rather than a measured result you can lean on.

What it cost when we called it (first-hand evidence)

Rate cards describe a model; a billed meter measures it. So we called GLM-5 through OpenRouter on 2026-07-10 and read the invoice. The served build came back as glm-5-20260211. A short prompt of 29 input tokens returned 150 output tokens in 7.62 seconds and billed $0.00050005. That is the priciest single run in our open-weight test set, and it is measured, not quoted from a rate card.

To keep the comparison honest we ran the same short prompt across the set the same day. Every number below is measured via OpenRouter routing, and because routing can price or latency differently from a native first-party endpoint, these carry a needs-native-reverify flag:

Two things stood out when we lined the runs up. First, GLM-5 was the slowest in the set at 7.62 seconds, against 2.8 seconds for the two Qwen runs on comparable prompts, so it is not a model you reach for when response latency is the constraint. Second, the billed cost was an order of magnitude above the cheapest run even though the prompts were similar in size, which tracks the published $0.60/$1.92 rate rather than hiding a surprise markup. We re-read the meter to be sure the $0.00050005 figure was the routed billed cost and not an estimate. The served string confirms we hit the February build, and we flag the routing because a native Zhipu-hosted call could differ; verify against the first-party endpoint before you budget production on these exact numbers.

Where GLM-5 fits for agent builders

The natural question is who should pay GLM-5's premium when Qwen3-235B-A22B exists at a twentieth of the output price. The honest answer is that price alone does not settle it, because these short latency-and-cost probes do not measure coding or agent quality, and GLM-5's whole pitch is quality on exactly those tasks.

According to Z.ai, the GLM line is built for programming and long-duration agent workloads, so the case for GLM-5 is that a stronger driver of a long agent loop can be worth more than a cheaper token if it finishes tasks in fewer steps or fewer retries. That is a real tradeoff, and it is the one buyers should actually run: a model that costs 19 times more per output token still wins on total cost if it needs a fraction of the turns. But that has to be proven on your own workload, not assumed from a rate card.

The verdict is narrow and decisive. Choose GLM-5 when your workload is a programming or long-horizon agent task where output quality per step matters more than raw token price, and you are willing to benchmark it against cheaper open weights on your own evals. Avoid it when your job is latency-sensitive, cost-dominated, or output-heavy at scale, where the 7.62-second run and the $1.92 output rate make Qwen3-235B-A22B or MiniMax-M2.5 the saner starting point until GLM-5 proves it earns the premium.

FAQ

What is GLM-5? GLM-5 is an open-release large language model from Zhipu AI (Z.ai) tuned for programming and long-duration agent tasks rather than general chat. On OpenRouter it is served as glm-5-20260211 and prices at roughly $0.60 input and $1.92 output per million tokens.

How much does GLM-5 cost to run? On OpenRouter it prices at about $0.60 per million input tokens and $1.92 per million output tokens. In our live test on 2026-07-10, a 29-input, 150-output call billed $0.00050005 and took 7.62 seconds, measured via OpenRouter routing.

Is GLM-5 expensive compared to other Chinese open-weight models? Yes, relatively. In our July 2026 test set GLM-5's $1.92 output rate was the highest, roughly 19 times Qwen3-235B-A22B's $0.10 and more than double MiniMax-M2.5's $0.90. Among cheap open-weight models, GLM-5 is the premium tier.

Why is GLM-5 slower than the other models you tested? On a comparable short prompt GLM-5 returned in 7.62 seconds, the slowest in our set, against about 2.8 seconds for the Qwen runs. We measured it via OpenRouter, so a native Zhipu endpoint could differ; the figure carries a needs-native-reverify flag.

Are these GLM-5 prices native first-party rates? No. Every price and latency here was measured via OpenRouter routing, not Zhipu's native endpoint, so they carry a needs-native-reverify flag. Confirm against the first-party rate before budgeting production spend on these exact figures.

This is part of the open-source AI agents in 2026 hub. For the full rate card across Zhipu's lineup, see the GLM API pricing hub.

Author: Kevin Fan, Customer Success Manager at China LLM Directory, specializing in Chinese LLM ecosystem pricing and model benchmarking. Last verified: 2026-07-10.