Up to date
753B parameters (8 of 256 experts per token), text in, text out, 1M tokens of context. 28 providers serve it from $0.034 per million input tokens.
Verified 2026-10-25 Rank #9 in LLMs · score 131.1 method v1 Sources 5 Downloads 30d 322.7k · #211 of 619 tracked
Changed this week: Cheapest input price fell 31.8%: $0.050 → $0.034 per 1M (OpenRouter).2026-10-24
No drawable SVG in the reply.
Why did the open-weights model cross the road?
Because it couldn't weight!
* .---.
/ \
/ \
\ /
\ /
| |
\ /
| |
_o_
/ | \
|
~~~~~~~~~_/\_~~~~~~~~~No playable frames in the reply.
The ringer: the same four playful tests for every model, replies shown as received and checked by code. A joke, not a benchmark; never ranked.2 of 4 pass · parasail-glm-52 on parasail · run 2026-10-09 · temperature 0 · cost $0.157 · 2026-10 thing: supermarket · all models →
Picked by rule from 35 tracked pieces: the lab's own post, the most shared, a field report, the latest.
GLM 5.2 lacks vision support, preventing it from reading image-based PDFs, screenshots, and design files.
ZCode is an official harness for GLM-5.3 offering 10,000 Credits per week and rolling access to flagship models.
colibri runs frontier MoE models like GLM-5.2 (744B parameters) on existing hardware by streaming experts from disk, using only 40B parameters per token with 11 GB of routed experts changing per token.
Z.ai released GLM-5.2, a 744B parameter open model with 40B active parameters and a 1M context window.
GLM 5.2 from Z.ai acts as a drop-in replacement for Opus and GPT 5.5, costing around $4.40/MTok which is less than 20% of the retail price of Opus.
The 6 picks above, plus these 19. 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.
27 providers with a live price, from $0.034 to $2.05 per million input tokens; the median is $0.97. One dot per provider, cheapest in teal.
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.
Widely served (28 API providers); heavily run locally (149 quantized builds).
Top Spaces running it: huggingface/ai-deadlines ♥ 783 · smolagents/ml-intern ♥ 423 · akhaliq/GLM-5.2 ♥ 11 · ICML-2026-agent-repro/logbook-judge ♥ 11
Weights only, an estimate: leave room for the KV cache. Method
About 452 GB at INT4, 904 GB at FP8, 1.81 TB 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
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-glm-52",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 4934,
"ttft_any_ms_median": 649,
"tokens_per_s_median": 230.7,
"runs": [
{
"ttft_s": 4.874,
"ttft_any_s": 0.767,
"tokens_per_s": 163.8,
"latency_s": 5.4,
"tokens_out": 884,
"reasoning_chars": 3388,
"finish": "stop"
},
{
"ttft_s": null,
"ttft_any_s": 0.524,
"tokens_per_s": 250.9,
"latency_s": 5.98,
"tokens_out": 1500,
"reasoning_chars": 5633,
"finish": "length"
},
{
"ttft_s": 4.993,
"ttft_any_s": 0.649,
"tokens_per_s": 230.7,
"latency_s": 5.52,
"tokens_out": 1274,
"reasoning_chars": 4008,
"finish": "stop"
}
]
}
],
"note": "Single client, wall clock; tok/s counts every output token including hidden reasoning."
} {
"calls": [
{
"name": "get_weather",
"args": {
"city": "Lisbon",
"date": "2026-10-14"
}
},
{
"name": "check_calendar",
"args": {
"date": "2026-10-14"
}
},
{
"name": "book_table",
"args": {
"restaurant": "Cervejaria Ramiro",
"date": "2026-10-14",
"time": "19:30",
"party_size": 2
}
}
],
"checks": {
"get_weather": true,
"check_calendar": true,
"book_table": true
},
"final": "Everything's set! Here's a summary:\n\n- **Weather**: Clear skies in Lisbon on Wednesday, 14 October 2026 — no rain expected (high of 22°C).\n- **Calendar**: Your free evening slot was **19:30–21:30**.\n- **Restaurant booking**: Table for 2 at **Cervejaria Ramiro** on **2026-10-14 at 19:30**.\n- **Confirmation code: TBL-4471**\n\nEnjoy your evening out! 🦐",
"fixture": "receipts.errand.v1"
} Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-06-16), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
176 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 |
|---|---|---|---|---|
| author-reported AIME 2026 · Reason through hard problems | 99.2 | — | model author | 2026-10-07 |
| author-reported CritPt | 20.9 | — | model author | 2026-10-07 |
| author-reported DeepSWE · Write code with agents | 46.2 | — | model author | 2026-10-07 |
| author-reported FrontierSWE · Write code with agents | 74.4 | — | model author | 2026-10-07 |
| author-reported GPQA Diamond · Reason through hard problems | 91.2 | — | model author | 2026-10-07 |
| author-reported HLE (w/ Tools) | 54.7 | — | model author | 2026-10-07 |
| author-reported HMMT · Reason through hard problems | 94.4 | — | model author | 2026-10-07 |
| author-reported HMMT · Reason through hard problems | 92.5 | — | model author | 2026-10-07 |
| author-reported Humanity's Last Exam · Reason through hard problems | 40.5 | — | model author | 2026-10-07 |
| author-reported IMO-AnswerBench · Reason through hard problems | 91 | — | model author | 2026-10-07 |
| author-reported MCP-Atlas (Public Set) | 76.8 | — | model author | 2026-10-07 |
| author-reported NL2Repo · Write code with agents | 48.9 | — | model author | 2026-10-07 |
| author-reported PostTrainBench | 34.3 | — | model author | 2026-10-07 |
| author-reported ProgramBench | 63.7 | — | model author | 2026-10-07 |
| author-reported SWE-Marathon | 13 | — | model author | 2026-10-07 |
| author-reported SWE-bench Pro · Write code with agents | 62.1 | — | model author | 2026-10-07 |
| author-reported Terminal-Bench · Write code with agents | 81 | — | model author | 2026-10-07 |
| author-reported Terminal-Bench · Write code with agents | 82.7 | — | model author | 2026-10-07 |
| author-reported Tool-Decathlon | 48.2 | — | model author | 2026-10-07 |
| author-reported deep_swe (datacurve/deep-swe) | 46.2 | — | Model Card | 2026-10-24 |
| author-reported diamond (Idavidrein/gpqa) | 91.2 | — | Model Card | 2026-10-24 |
| author-reported hle (cais/hle) | 40.5 | — | Model Card | 2026-10-24 |
| author-reported hle (cais/hle) | 54.7 | With tools | Model Card | 2026-10-24 |
| author-reported SWE_Bench_Pro (ScaleAI/SWE-bench_Pro) | 62.1 | — | Model Card | 2026-10-24 |
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 | GlmMoeDsaForCausalLM |
| Model type | glm_moe_dsa |
| Dtype | bfloat16 |
| Params by dtype | BF16: 753,329,921,024 · F32: 19,456 |
| Size band | XXL |
| MoE | yes |
| Experts | 256 |
| Experts per token | 8 |
| Context length | 1,048,576 (config.json max_position_embeddings) |
| Languages | en, zh |
| Papers | arXiv:2602.15763, arXiv:2603.12201 |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · GlmMoeDsaForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: not in registry · 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-5.2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.81 TB · Cluster (≥640 GB) |
|---|---|
| FP8 (1 byte/param) | 904 GB · Cluster (≥640 GB) |
| INT4 (0.5 bytes/param) | 452 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: 149finetune: 26adapter: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| nvidia/GLM-5.2-NVFP4 | quantized | NVFP4 | 381B | XXL | mit | nvidia | 328 | 3.9M |
| unsloth/GLM-5.2-GGUF | quantized | GGUF | — | — | mit | unsloth | 647 | 1.4M |
| cyankiwi/GLM-5.2-AWQ-INT4 | quantized | AWQ | 753B | XXL | mit | cyankiwi | 15 | 519.8k |
| amd/GLM-5.2-MXFP4 | quantized | MXFP4 | 412B | XXL | mit | amd | 18 | 307.2k |
| local-inference-lab/GLM-5.2-NVFP4 | quantized | NVFP4 | 386B | XXL | mit | local-inference-lab | 32 | 123.1k |
| brandonmusic/GLM-5.2-EXL3-TR3v4-3.5bpw-MTP78 | quantized | EXL3 | 173B | XXL | mit | brandonmusic | 9 | 101k |
| QuantTrio/GLM-5.2-Int4-Int8Mix | quantized | Int4 | 785B | XXL | mit | QuantTrio | 21 | 100k |
| pipenetwork/GLM-5.2-REAP50-Q3_K_M-GGUF | quantized | GGUF | — | — | mit | pipenetwork | 15 | 87.8k |
| huihui-ai/Huihui-GLM-5.2-abliterated-GGUF | quantized | GGUF | — | — | mit | huihui-ai | 360 | 50.4k |
| sokann/GLM-5.2-GGUF-2.244bpw | quantized | GGUF | — | — | mit | sokann | 26 | 35.7k |
| RedHatAI/GLM-5.2-NVFP4-FP8 | quantized | NVFP4 | 425B | XXL | mit | RedHatAI | 25 | 35.6k |
| malaiwah/GLM-5.2-EXL3-FQ-segments | quantized | EXL3 | — | — | mit | malaiwah | 0 | 35.2k |
| mlx-community/GLM-5.2-DQ4plus-q8 | quantized | — | 743B | XXL | mit | mlx-community | 18 | 32.9k |
| mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp | quantized | int4 | — | — | mit | mastouri | 144 | 25.4k |
| 0xSero/GLM-5.2-504B | quantized | — | 290B | XXL | mit | 0xSero | 33 | 25k |
| baseten/GLM-5.2-Vision-NVFP4 | quantized | NVFP4 | 381B | XXL | mit | baseten | 155 | 18.6k |
| pipenetwork/GLM-5.2-MLX-mixed-3_6bit | quantized | MLX | 743B | XXL | mit | pipenetwork | 6 | 21.3k |
| mlx-community/GLM-5.2-4bit | quantized | -4bit | 743B | XXL | mit | mlx-community | 6 | 19.2k |
| modal-labs/glm-5.2-nvfp4-experts-only | quantized | nvfp4 | 381B | XXL | other | modal-labs | 0 | 19.5k |
| 0xSero/GLM-5.2-REAP-504B-GGUF | quantized | GGUF | — | — | mit | 0xSero | 48 | 16.1k |
| canada-quant/GLM-5.2-W4A16-MTP | quantized | -W4A16 | 116B | XXL | mit | canada-quant | 21 | 16.8k |
| inferencerlabs/GLM-5.2-MLX-Q4i | quantized | MLX | — | — | — | inferencerlabs | 9 | 16.9k |
| inference-optimization/GLM-5.2-0.8B-A0.8B | finetune | — | 850M | S | mit | inference-optimization | 4 | 16.3k |
| philipjohnbasile/GLM-5.2-Demolition-q4a4-soul-MLX | quantized | MLX | 186B | XXL | mit | philipjohnbasile | 12 | 15.6k |
| kernelpool/GLM-5.2-8bit | quantized | -8bit | 743B | XXL | mit | kernelpool | 0 | 15.5k |
| avlp12/GLM-5.2-Alis-MLX-Dynamic-3.5bpw | quantized | MLX | 751B | XXL | mit | avlp12 | 5 | 14.7k |
| SixVolts/GLM-5.2-ewaste-edition-GGUF | quantized | GGUF | — | — | mit | SixVolts | 15 | 13.3k |
| Mapika/GLM-5.2-NVFP4 | quantized | NVFP4 | 379B | XXL | mit | Mapika | 7 | 12.3k |
| pipenetwork/GLM-5.2-MLX-4bit | quantized | MLX | 743B | XXL | mit | pipenetwork | 3 | 11.9k |
| inferencerlabs/GLM-5.2-MTP-MLX | finetune | MLX | 9.95B | L | — | inferencerlabs | 7 | 11.6k |
| pipenetwork/GLM-5.2-MLX-8bit | quantized | MLX | 743B | XXL | mit | pipenetwork | 0 | 11.7k |
| avlp12/GLM-5.2-Alis-MLX-Dynamic-2.56bpw | quantized | MLX | 751B | XXL | mit | avlp12 | 2 | 10.4k |
| DevQuasar/zai-org.GLM-5.2-GGUF | quantized | GGUF | — | — | — | DevQuasar | 3 | 10.2k |
| muzzy/GLM-5.2-GGUF | quantized | GGUF | — | — | mit | muzzy | 9 | 8.9k |
| tiny-random/glm-5.2 | finetune | — | 13.2M | S | — | tiny-random | 1 | 8.9k |
| spicyneuron/GLM-5.2-MLX-4.5bit | quantized | MLX | 743B | XXL | mit | spicyneuron | 2 | 8.3k |
| bearzi/GLM-5.2-JANGTQ_K | finetune | — | 79.2B | XL | mit | bearzi | 3 | 8.2k |
| pipenetwork/GLM-5.2-REAP50-Q2_K-GGUF | quantized | Q2_K | — | — | mit | pipenetwork | 3 | 7.6k |
| unsloth/GLM-5.2-FP8 | quantized | -FP8 | 753B | XXL | mit | unsloth | 11 | 6.8k |
| lowbitcoffee/GLM-5.2-W4A16 | quantized | -W4A16 | 101B | XXL | mit | lowbitcoffee | 4 | 7k |
| Date | Commit | SHA |
|---|---|---|
| 2026-09-01 | Update README.md | cf457fa |
| 2026-07-02 | add moe_router_dtype config | b4734de |
| 2026-06-23 | add Footnote | f226310 |
| 2026-06-22 | Add community evaluation results for DEEP-SWE, GPQA, HLE, SWE-BENCH_PRO (#12) | e32aaf0 |
| 2026-06-19 | add unsloth | 5378302 |
| 2026-06-17 | Update README.md | 4d67f66 |
| 2026-06-17 | update with Ascend support | 120edc2 |
| 2026-06-16 | update readme | f6142f1 |
| 2026-06-16 | update readme | b4d667e |
| 2026-06-16 | init readme | 48cf768 |
| 2026-06-16 | Add files using upload-large-folder tool | 96b232d |
| 2026-06-16 | Add files using upload-large-folder tool | f3ce794 |
| 2026-06-16 | Add files using upload-large-folder tool | a66456a |
| 2026-06-16 | Add files using upload-large-folder tool | 08c85fa |
| 2026-06-16 | initial commit | b416619 |
Hosted endpoints observed for GLM-5.2, 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: $0.03/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Wafer openrouter | z-ai/glm-5.2 | 1M | fp4 | $0.03 | $7.00 | — | — | yes | live | 2026-10-24 |
| Baidu openrouter | z-ai/glm-5.2 | 1M | fp8 | $0.19 | $0.59 | — | — | yes | live | 2026-10-24 |
| Cloudflare openrouter | z-ai/glm-5.2 | 256K | unknown | $0.19 | $8.00 | — | — | yes | live | 2026-10-24 |
| StreamLake openrouter | z-ai/glm-5.2 | 1000K | fp8 | $0.28 | $0.88 | — | — | yes | live | 2026-10-24 |
| Relace openrouter | z-ai/glm-5.2 | 1M | unknown | $0.40 | $4.00 | — | — | yes | live | 2026-10-24 |
| DeepInfra hf, deepinfra, openrouter | zai-org/GLM-5.2 | 1M | fp4 | $0.45 | $1.44 | 70 | 620 ms | yes | live | 2026-10-25 |
| Morph openrouter | z-ai/glm-5.2 | 1M | fp8 | $0.45 | $4.40 | — | — | yes | live | 2026-10-24 |
| Ambient openrouter | z-ai/glm-5.2 | 198K | fp8 | $0.60 | $2.00 | — | — | yes | live | 2026-10-24 |
| CoreWeave openrouter | z-ai/glm-5.2 | 1M | fp4 | $0.62 | $2.42 | — | — | yes | live | 2026-10-24 |
| Novita hf, openrouter | zai-org/glm-5.2 | 1M | fp8 | $0.65 | $2.04 | 51.1 | 1813 ms | yes | live | 2026-10-25 |
| DigitalOcean openrouter | z-ai/glm-5.2 | 1M | unknown | $0.70 | $2.20 | — | — | yes | live | 2026-10-24 |
| SiliconFlow openrouter | z-ai/glm-5.2 | 1M | fp8 | $0.70 | $2.20 | — | — | yes | live | 2026-10-24 |
| AtlasCloud openrouter | z-ai/glm-5.2 | 1M | fp8 | $0.94 | $2.95 | — | — | yes | live | 2026-10-24 |
| Alibaba openrouter | z-ai/glm-5.2 | 977K | fp8 | $0.97 | $3.04 | — | — | yes | live | 2026-10-24 |
| Inceptron openrouter | z-ai/glm-5.2 | 1M | fp4 | $0.99 | $1.19 | — | — | yes | live | 2026-10-24 |
| Decart openrouter | z-ai/glm-5.2 | 1M | fp4 | $1.20 | $8.00 | — | — | yes | live | 2026-10-24 |
| Phala openrouter | z-ai/glm-5.2 | 1M | fp8 | $1.26 | $3.00 | — | — | yes | live | 2026-10-24 |
| Together AI hf, openrouter | zai-org/GLM-5.2 | 1024K | unknown | $1.40 | $4.40 | 200.5 | 224 ms | yes | live | 2026-10-25 |
| Z.ai hf, openrouter | glm-5.2 | 1M | fp8 | $1.40 | $4.40 | 56 | 2691 ms | yes | live | 2026-10-25 |
| Baseten hf, openrouter | zai-org/GLM-5.2 | 1M | fp8 | $1.40 | $4.40 | 87.5 | 1004 ms | yes | live | 2026-10-25 |
| Parasail parasail, openrouter | parasail-glm-52 | 256K | fp4 | $1.40 | $4.40 | — | — | yes | serverless | 2026-10-24 |
| Nebius openrouter | z-ai/glm-5.2 | 1M | fp4 | $1.40 | $4.40 | — | — | yes | live | 2026-10-24 |
| Mistral openrouter | z-ai/glm-5.2 | 1M | nvfp4 | $1.40 | $4.40 | — | — | yes | live | 2026-10-24 |
| Venice openrouter | z-ai/glm-5.2 | 977K | fp8 | $1.40 | $4.40 | — | — | yes | live | 2026-10-24 |
| GMICloud openrouter | z-ai/glm-5.2 | 1M | fp8 | $1.40 | $4.40 | — | — | yes | live | 2026-10-24 |
| Friendli openrouter | z-ai/glm-5.2 | 1M | unknown | $1.40 | $4.40 | — | — | yes | live | 2026-10-24 |
| Scaleway hf | glm-5.2 | — | — | $2.05 | $6.27 | 138.9 | 481 ms | yes | live | 2026-10-25 |
| Featherless hf | zai-org/GLM-5.2 | — | — | — | — | — | — | — | live | 2026-10-25 |
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 21:31 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 14:59 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.