Up to date
358B parameters (8 of 160 experts per token), text in, text out, 128K tokens of context. 1 provider serves it from $0.60 per million input tokens.
Verified 2026-10-25 Rank #150 in LLMs · score 93.1 method v1 Sources 4 Downloads 30d 113.8k · #305 of 619 tracked
Picked by rule from 20 tracked pieces: the lab's own post, the most shared, a field report, the latest.
GLM-4.5 evaluation for SWE-bench Verified uses OpenHands v0.34.0 with runs limited to 100 iterations.
Z.ai introduces GLM-4.5 and GLM-4.5-Air, hybrid reasoning models designed to unify reasoning, coding, and agentic capabilities.
GLM-4.5 Air, a 106 billion parameter MIT licensed model from Z.ai, runs locally on a 64GB MacBook Pro M2 via a 44GB 3bit quantized version.
Z.AI released GLM-4.5 and GLM-4.5-Air, foundational models optimized for agent-oriented applications using a Mixture-of-Experts architecture.
quantprobe predicts token speeds and memory fit for GGUF models on specific hardware without downloading them first.
The 4 picks above, plus these 16. Top is the radar's score: discussion, source authority and the lab's own posts, fading with age.
Live hosted endpoints, cheapest first, USD per million tokens. Ties break on output price, then provider name.
| # | Cheapest live row | In /1M | Out /1M | Context | Quant | Seen |
|---|---|---|---|---|---|---|
| 1 | Z.AI | $0.60 | $2.20 | 128K | fp8 | Oct 24 |
Hugging Face does not say who downloads a model or with what. These are the proxies: API providers listing it, quantized builds, fine-tunes/adapters/merges, and Spaces that load it.
Served by 1 API provider; 26 quantized builds for local use.
Top Spaces running it: zai-org/GLM-4.5-Space ♥ 110 · jairwaal/image ♥ 16 · Tigasturned/GLM-4.5-WebDev ♥ 10 · qgallouedec/tito ♥ 7
Weights only, an estimate: leave room for the KV cache. Method
About 215 GB at INT4, 430 GB at FP8, 860 GB at BF16: weights only, so leave room for the KV cache. Every expert must be resident, so total parameters count.
Estimate params × bytes per param × 1.2 · how we estimate
Third-party measurements only; author-reported numbers stay in Reference, marked as such.
64.2. Each tick is a model on this site with a measured SWE-bench Verified row; author-reported numbers stay in Reference. Percent resolved. agent: Undisclosed
Source · leaderboard · 2025-07-28
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-07-20), amber triangles are 8 dated articles (each links to it), grey ticks are repo commits.
34 derivatives on the Hub. Which build to run, by runtime: the most downloaded in each class.
From the Hugging Face model tree; quantized builds in teal, fine-tunes, merges and adapters in greys.
Everything else, folded: the full tables, every source.
| Benchmark | Value | Setting / notes | Reported by | Date |
|---|---|---|---|---|
| measured SWE-bench Verified · Write code with agents | 64.2 percent resolved | agent: Undisclosed | leaderboard | 2025-07-28 |
| measured SWE-bench Verified · Write code with agents | 54.2 percent resolved | agent: mini-SWE-agent | leaderboard | 2025-08-22 |
| author-reported comprehensive evaluation across 12 industry-standard benchmarks | 63.2 | GLM-4.5 | model author | 2026-10-07 |
| author-reported comprehensive evaluation across 12 industry-standard benchmarks | 59.8 | GLM-4.5-Air | model author | 2026-10-07 |
Labels are evidence, not verdicts: measured a third-party leaderboard row for this model; author-reported a number the model author published in the card or eval files; structural a deterministic signal from the name, config or chat template, or a decision-model estimate. Each chip links to its source. All use cases.
| Pipeline tag | text-generation |
|---|---|
| Library | transformers |
| Architectures | Glm4MoeForCausalLM |
| Model type | glm4_moe |
| Dtype | bfloat16 |
| Params by dtype | BF16: 358,337,776,896 · F32: 14,400 |
| Size band | XXL |
| MoE | yes |
| Experts | 160 |
| Experts per token | 8 |
| Context length | 131,072 (config.json max_position_embeddings) |
| Languages | en, zh |
| Papers | arXiv:2508.06471 |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Glm4MoeForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Glm4MoeForCausalLM · 773cb938fe9484a8f7e7e46fb1b1fe96412c59ed
vLLM registry (7050ef881ff0835f6890a47db1859a24c65481d6) · SGLang registry (773cb938fe9484a8f7e7e46fb1b1fe96412c59ed) · Matched by architecture class name against each project's registry on main; 'supported' means the architecture is registered, not that this checkpoint was tested.
Weights-only memory for GLM-4.5: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 860 GB · Cluster (≥640 GB) |
|---|---|
| FP8 (1 byte/param) | 430 GB · Multi-GPU node (<640 GB) |
| INT4 (0.5 bytes/param) | 215 GB · Multi-GPU node (<640 GB) |
Weights only (params × bytes/param × 1.2 overhead); KV cache and activations extra. MoE: all experts must be resident, so total parameters count.
quantized: 26finetune: 7merge: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| unsloth/GLM-4.5-GGUF | quantized | GGUF | — | — | mit | unsloth | 52 | 91.4k |
| tiny-random/glm-4-moe | finetune | — | 2.57M | S | — | tiny-random | 2 | 26.6k |
| bartowski/zai-org_GLM-4.5-GGUF | quantized | GGUF | — | — | — | bartowski | 8 | 19.9k |
| ubergarm/GLM-4.5-GGUF | quantized | GGUF | — | — | mit | ubergarm | 21 | 14k |
| mlx-community/GLM-4.5-4bit | quantized | -4bit | 353B | XXL | mit | mlx-community | 16 | 7.5k |
| QuantTrio/GLM-4.5-AWQ | quantized | AWQ | 52.6B | XL | — | QuantTrio | 9 | 4.6k |
| yujiepan/glm-4.5-tiny-random | finetune | — | 2.57M | S | — | yujiepan | 2 | 4k |
| yujiepan/glm-4-moe-tiny-random | finetune | — | 2.57M | S | — | yujiepan | 0 | 4k |
| DevQuasar/zai-org.GLM-4.5-GGUF | quantized | GGUF | — | — | — | DevQuasar | 2 | 3.9k |
| lmstudio-community/GLM-4.5-GGUF | quantized | GGUF | — | — | — | lmstudio-community | 0 | 3.9k |
| QuantTrio/GLM-4.5-GPTQ-Int4-Int8Mix | quantized | GPTQ | 55.3B | XL | — | QuantTrio | 4 | 2k |
| anikifoss/GLM-4.5-HQ4_K | quantized | GGUF | — | — | mit | anikifoss | 1 | 1.4k |
| lovedheart/GLM-4.5-GGUF-IQ1_M | quantized | GGUF | — | — | mit | lovedheart | 2 | 1.1k |
| unsloth/GLM-4.5 | finetune | — | 358B | XXL | mit | unsloth | 7 | 391 |
| sjug/GLM-4.5-MLX-9bit | quantized | MLX | 353B | XXL | mit | sjug | 0 | 573 |
| sm54/GLM-4.5-MXFP4_MOE | quantized | GGUF | — | — | — | sm54 | 2 | 432 |
| brAInwav/GLM-4.5-mlx-4Bit | quantized | mlx | 353B | XXL | mit | brAInwav | 0 | 484 |
| monirmamoun/GLM-4.5-MLX-3bit | quantized | MLX | 353B | XXL | mit | monirmamoun | 0 | 471 |
| sokann/GLM-4.5-GGUF-3.2162bpw | quantized | GGUF | — | — | mit | sokann | 0 | 440 |
| monirmamoun/GLM-4.5-MLX-2bit | quantized | MLX | 353B | XXL | mit | monirmamoun | 0 | 431 |
| tiny-random/glm-4.5 | finetune | — | 2.57M | S | — | tiny-random | 2 | 257 |
| finding1/GLM-4.5-MLX-8.5bpw | quantized | MLX | 353B | XXL | mit | finding1 | 0 | 341 |
| mradermacher/GLM-4.5-i1-GGUF | quantized | GGUF | — | — | mit | mradermacher | 0 | 248 |
| bullerwins/GLM-4.5-exl3-2.76bpw_optim | quantized | exl3 | 61.7B | XL | mit | bullerwins | 1 | 126 |
| bullerwins/GLM-4.5-exl3-2.75bpw | quantized | exl3 | 61.8B | XL | mit | bullerwins | 1 | 119 |
| manaladan6/zai-org.GLM-4.5-GGUF | quantized | GGUF | — | — | — | manaladan6 | 0 | 149 |
| bullerwins/GLM-4.5-exl3-3.2bpw_optim | quantized | exl3 | 71.4B | XL | mit | bullerwins | 0 | 114 |
| bullerwins/GLM-4.5-exl3-5.0bpw | quantized | exl3 | 111B | XXL | mit | bullerwins | 0 | 112 |
| bullerwins/GLM-4.5-exl3-4.0bpw | quantized | exl3 | 89.2B | XL | mit | bullerwins | 0 | 109 |
| bullerwins/GLM-4.5-exl3-3.0bpw | quantized | exl3 | 67.3B | XL | mit | bullerwins | 0 | 108 |
| bullerwins/GLM-4.5-exl3-3.5bpw | quantized | exl3 | 78.3B | XL | mit | bullerwins | 0 | 99 |
| mradermacher/GLM-4.5-GGUF | finetune | GGUF | — | — | mit | mradermacher | 1 | 0 |
| DINESH5578/Razerbills-Advanced-researcher | merge | — | — | — | apache-2.0 | DINESH5578 | 0 | 0 |
| Vasanth6353/Geetha | finetune | — | — | — | apache-2.0 | Vasanth6353 | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2025-08-11 | Improve model card with detailed quick start, system requirements, and paper link clarification (#20) | cbb2c7c |
| 2025-08-11 | tc update | e8b2a04 |
| 2025-07-28 | global maas api link | 9cfe10c |
| 2025-07-28 | update | 8b91a96 |
| 2025-07-28 | update | fa9c14b |
| 2025-07-28 | update final | a0ea464 |
| 2025-07-28 | update | 1f7bf97 |
| 2025-07-28 | upload | 51f1156 |
| 2025-07-28 | upload | 8ce387f |
| 2025-07-28 | Add files using upload-large-folder tool | 9df55da |
| 2025-07-28 | Add files using upload-large-folder tool | 2a2aec4 |
| 2025-07-20 | initial commit | c066022 |
Hosted endpoints observed for GLM-4.5, sorted by input price (USD per million tokens). Each row names its observer and date; prices change daily, so check the source before quoting. Cheapest observed input price: $0.60/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Z.AI openrouter | z-ai/glm-4.5 | 128K | fp8 | $0.60 | $2.20 | — | — | yes | live | 2026-10-24 |
Ranked by source signals (HN, Lobsters, authority domains, official posts) and recency. Summaries are generated from a stored copy of the article text; every sentence cites a span of that text, and nothing is shown that failed the citation check.
arXiv links found in the model card and its tags. "Peer-reviewed" is shown only with evidence: a published venue in OpenAlex (measured) or the authors' own journal reference on arXiv (reported).
Hub snapshot: 2026-10-25 · HF API: 2026-10-24 21:34 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:01 UTC · Last verified 2026-10-25
Email alerts are not open yet; the feeds above are the same data. See feeds for the item schema.
