Up to date
Qwen Apache 2.0 out 6 months ago
36.0B parameters, 3B active (8 of 256 experts per token), images and text in, text out, 256K tokens of context. 12 providers serve it from $0.050 per million input tokens.
Verified 2026-10-25 Rank #15 in LLMs · score 121.8 method v1 Sources 5 Downloads 30d 3.9M · #37 of 619 tracked
Changed this week: 32 new quantized builds on the Hub (918 in all).2026-10-24 · 29 new quantized builds on the Hub (915 in all).2026-10-24
Why did the open-weights model cross the road?
To get to the other fine-tune!
/ \ /___\ | | | 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.3 of 4 pass · parasail-qwen3p6-35b-a3b on parasail · run 2026-10-09 · temperature 0 · cost $0.022 · 2026-10 thing: supermarket · all models →
Picked by rule from 37 tracked pieces: the lab's own post, the most shared, a field report, the latest.
Qwen3.6-35B-A3B is a sparse mixture-of-experts model with 35 billion total parameters and only 3 billion active parameters.
Qwen3.6-35B-A3B running on a laptop generated a better pelican illustration than Claude Opus 4.7.
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.
11 providers with a live price, from $0.050 to $0.28 per million input tokens; the median is $0.10. 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 (12 API providers); heavily run locally (918 quantized builds); a big base for builders (790 fine-tunes, adapters and merges).
Top Spaces running it: embedl/hfviewer ♥ 181 · KyleHessling1/qwopus36-35b-a3b-eval ♥ 12 · emanforhelp/model-pulse ♥ 5 · curtburk/vLLM_on_NVIDIA_GB10 ♥ 4
Weights only, an estimate: leave room for the KV cache. Method
About 22 GB at INT4, 43 GB at FP8, 86 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
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-qwen3p6-35b-a3b",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 13700,
"ttft_any_ms_median": 992,
"tokens_per_s_median": 114.4,
"runs": [
{
"ttft_s": null,
"ttft_any_s": 0.992,
"tokens_per_s": 114.4,
"latency_s": 13.11,
"tokens_out": 1500,
"reasoning_chars": 6374,
"finish": "length"
},
{
"ttft_s": 12.293,
"ttft_any_s": 0.849,
"tokens_per_s": 119.9,
"latency_s": 12.51,
"tokens_out": 1500,
"reasoning_chars": 7042,
"finish": "length"
},
{
"ttft_s": 15.106,
"ttft_any_s": 1.098,
"tokens_per_s": 92.1,
"latency_s": 16.29,
"tokens_out": 1500,
"reasoning_chars": 6902,
"finish": "length"
}
]
}
],
"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": "The weather in Lisbon on Wednesday, October 14, 2026, is clear with a very low chance of rain. I found a free evening slot in your calendar from 19:30 to 21:30 and successfully booked a table for 2 at Cervejaria Ramiro at 19:30.\n\nYour confirmation code is: **TBL-4471**",
"fixture": "receipts.errand.v1"
} Weights, every dated article, and repo commits on one rail.
Teal is the weights (2026-04-15), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
1,708 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 NL2Repo · Write code with agents | 29.4 | — | model author | 2026-10-07 |
| author-reported QwenClawBench | 52.6 | — | model author | 2026-10-07 |
| author-reported QwenWebBench | 1397 | — | model author | 2026-10-07 |
| author-reported SWE-bench Multilingual · Write code with agents | 67.2 | — | model author | 2026-10-07 |
| author-reported SWE-bench Pro · Write code with agents | 49.5 | — | model author | 2026-10-07 |
| author-reported SWE-bench Verified · Write code with agents | 73.4 | — | model author | 2026-10-07 |
| author-reported TAU3-Bench | 67.2 | — | model author | 2026-10-07 |
| author-reported Terminal-Bench · Write code with agents | 51.5 | — | model author | 2026-10-07 |
| author-reported VITA-Bench | 35.6 | — | 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 | image-text-to-text |
|---|---|
| Library | transformers |
| Architectures | Qwen3_5MoeForConditionalGeneration |
| Model type | qwen3_5_moe |
| Dtype | bfloat16 |
| Params by dtype | BF16: 35,951,822,704 |
| Size band | XL |
| MoE | yes |
| Experts | 256 |
| Experts per token | 8 |
| Context length | 262,144 (config.json max_position_embeddings) |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Qwen3_5MoeForConditionalGeneration · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3_5MoeForConditionalGeneration · 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 Qwen3.6-35B-A3B: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 86 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 43 GB · Single high-end GPU (<80 GB) |
| INT4 (0.5 bytes/param) | 22 GB · Single GPU (<24 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: 918finetune: 395adapter: 357merge: 38
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Qwen/Qwen3.6-35B-A3B-FP8 | quantized | -FP8 | 36.0B | XL | apache-2.0 | Qwen | 438 | 47.2M |
| nvidia/Qwen3.6-35B-A3B-NVFP4 | quantized | NVFP4 | 18.7B | L | apache-2.0 | nvidia | 670 | 39.6M |
| HauhauCS/Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive | quantized | GGUF | — | — | apache-2.0 | HauhauCS | 4k | 12.4M |
| RedHatAI/Qwen3.6-35B-A3B-NVFP4 | quantized | NVFP4 | 34.7B | L | apache-2.0 | RedHatAI | 193 | 11.5M |
| unsloth/Qwen3.6-35B-A3B-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 1.7k | 9.7M |
| cyankiwi/Qwen3.6-35B-A3B-AWQ-4bit | quantized | AWQ | 36.0B | XL | apache-2.0 | cyankiwi | 96 | 6.2M |
| unsloth/Qwen3.6-35B-A3B-NVFP4 | quantized | NVFP4 | 23.0B | L | apache-2.0 | unsloth | 123 | 5.7M |
| unsloth/Qwen3.6-35B-A3B-MTP-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 1k | 5M |
| QuantTrio/Qwen3.6-35B-A3B-AWQ | quantized | AWQ | 36.0B | XL | apache-2.0 | QuantTrio | 37 | 4.9M |
| Bahushruth/Qwen3.6-35B-A3B-abliterated-v4 | finetune | — | 34.7B | L | apache-2.0 | Bahushruth | 13 | 3M |
| lmstudio-community/Qwen3.6-35B-A3B-MLX-4bit | quantized | MLX | 35.1B | XL | apache-2.0 | lmstudio-community | 17 | 2.2M |
| lmstudio-community/Qwen3.6-35B-A3B-GGUF | quantized | GGUF | — | — | apache-2.0 | lmstudio-community | 33 | 2.1M |
| lmstudio-community/Qwen3.6-35B-A3B-MLX-8bit | quantized | MLX | 35.1B | XL | apache-2.0 | lmstudio-community | 5 | 2.1M |
| lmstudio-community/Qwen3.6-35B-A3B-MLX-6bit | quantized | MLX | 35.1B | XL | apache-2.0 | lmstudio-community | 4 | 2M |
| unsloth/Qwen3.6-35B-A3B-NVFP4-Fast | quantized | NVFP4 | 19.8B | L | apache-2.0 | unsloth | 129 | 2M |
| palmfuture/Qwen3.6-35B-A3B-GPTQ-Int4 | quantized | GPTQ | 36.0B | XL | apache-2.0 | palmfuture | 31 | 1.8M |
| bartowski/Qwen_Qwen3.6-35B-A3B-GGUF | quantized | GGUF | — | — | apache-2.0 | bartowski | 167 | 1.5M |
| michaelw9999/Qwen3.6-35B-A3B-NVFP4-MTP-GGUF | quantized | NVFP4 | — | — | — | michaelw9999 | 11 | 1.1M |
| mudler/Qwen3.6-35B-A3B-APEX-GGUF | quantized | GGUF | — | — | apache-2.0 | mudler | 205 | 926.1k |
| z-lab/Qwen3.6-35B-A3B-DFlash | finetune | — | 386M | S | apache-2.0 | z-lab | 295 | 867.6k |
| ykarout/Qwen3.6-35B-A3B-NVFP4 | quantized | NVFP4 | 36.0B | XL | apache-2.0 | ykarout | 1 | 774.6k |
| ggml-org/Qwen3.6-35B-A3B-GGUF | quantized | GGUF | — | — | apache-2.0 | ggml-org | 28 | 732.9k |
| huihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated-MTP-GGUF | quantized | GGUF | — | — | apache-2.0 | huihui-ai | 57 | 708.8k |
| Intel/Qwen3.6-35B-A3B-int4-mixed-AutoRound | quantized | int4 | 6.33B | M | — | Intel | 13 | 553.2k |
| mlx-community/Qwen3.6-35B-A3B-4bit | quantized | -4bit | 35.1B | XL | apache-2.0 | mlx-community | 118 | 417.5k |
| rdtand/Qwen3.6-35B-A3B-PrismaQuant-4.75bit-vllm | quantized | — | 21.1B | L | apache-2.0 | rdtand | 35 | 421.2k |
| Ex0bit/Qwen3.6-35B-A3B-PRISM-NVFP4 | quantized | NVFP4 | — | — | apache-2.0 | Ex0bit | 22 | 360.7k |
| unsloth/Qwen3.6-35B-A3B-UD-MLX-4bit | quantized | MLX | 35.1B | XL | apache-2.0 | unsloth | 124 | 327.6k |
| flywheel-ai/automotive | quantized | GGUF | 34.7B | L | apache-2.0 | flywheel-ai | 0 | 324.1k |
| unsloth/Qwen3.6-35B-A3B | finetune | — | 36.0B | XL | apache-2.0 | unsloth | 36 | 281.8k |
| mlx-community/Qwen3.6-35B-A3B-8bit | quantized | -8bit | 35.1B | XL | apache-2.0 | mlx-community | 33 | 268.3k |
| lordx64/Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled | adapter | — | 36.0B | XL | apache-2.0 | lordx64 | 204 | 234.1k |
| mudler/Qwen3.6-35B-A3B-APEX-MTP-GGUF | quantized | GGUF | — | — | apache-2.0 | mudler | 76 | 206.1k |
| sakamakismile/Qwen3.6-35B-A3B-NVFP4 | quantized | NVFP4 | — | — | apache-2.0 | sakamakismile | 14 | 205.3k |
| Edge0/Edge0-35B-A3B-preview | adapter | — | 34.7B | L | apache-2.0 | Edge0 | 1.1k | 84.2k |
| selode-ai/Qwen-3.6-35B-A3B-VRAP-4-bit-AWQ-21.2GB | quantized | AWQ | 28.7B | L | apache-2.0 | selode-ai | 17 | 136.2k |
| lokeshe09/Qwen3.6-35B-A3B-INT8 | quantized | INT8 | 34.7B | L | apache-2.0 | lokeshe09 | 0 | 110.6k |
| prithivMLmods/Qwen3.6-35B-A3B-MTP-GGUF | quantized | GGUF | — | — | apache-2.0 | prithivMLmods | 3 | 106.9k |
| huihui-ai/Huihui-Qwen3.6-35B-A3B-abliterated | finetune | — | 36.0B | XL | apache-2.0 | huihui-ai | 101 | 88.4k |
| llmfan46/Qwen3.6-35B-A3B-uncensored-heretic-Native-MTP-Preserved | finetune | — | 35.1B | XL | apache-2.0 | llmfan46 | 47 | 90.8k |
Hosted endpoints observed for Qwen3.6-35B-A3B, 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: $0.05/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Darkbloom openrouter | qwen/qwen3.6-35b-a3b | 256K | fp4 | $0.05 | $0.70 | — | — | yes | live | 2026-10-24 |
| DeepInfra hf, deepinfra, openrouter | Qwen/Qwen3.6-35B-A3B | 256K | fp8 | $0.10 | $0.95 | 60.1 | 265 ms | yes | live | 2026-10-25 |
| Parasail parasail, openrouter | parasail-qwen3p6-35b-a3b | 256K | fp8 | $0.10 | $0.95 | — | — | yes | serverless | 2026-10-24 |
| AkashML openrouter | qwen/qwen3.6-35b-a3b | 256K | fp8 | $0.10 | $0.90 | — | — | yes | live | 2026-10-24 |
| DekaLLM openrouter | qwen/qwen3.6-35b-a3b | 256K | unknown | $0.10 | $1.00 | — | — | yes | live | 2026-10-24 |
| Venice openrouter | qwen/qwen3.6-35b-a3b | 250K | fp8 | $0.10 | $1.00 | — | — | yes | live | 2026-10-24 |
| Phala openrouter | qwen/qwen3.6-35b-a3b | 256K | unknown | $0.20 | $1.27 | — | — | yes | live | 2026-10-24 |
| SiliconFlow openrouter | qwen/qwen3.6-35b-a3b | 256K | fp8 | $0.24 | $1.80 | — | — | — | live | 2026-10-24 |
| CoreWeave openrouter | qwen/qwen3.6-35b-a3b | 256K | fp8 | $0.25 | $1.25 | — | — | — | live | 2026-10-24 |
| AtlasCloud openrouter | qwen/qwen3.6-35b-a3b | 256K | fp8 | $0.25 | $1.25 | — | — | yes | live | 2026-10-24 |
| Scaleway hf | qwen3.6-35b-a3b | — | — | $0.28 | $1.71 | 105.7 | 547 ms | yes | live | 2026-10-25 |
| Featherless hf | Qwen/Qwen3.6-35B-A3B | — | — | — | — | — | — | — | 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.
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.