Up to date
Qwen Apache 2.0 out 15 months ago
235B parameters, 22B active (8 of 128 experts per token), text in, text out, 256K tokens of context. 7 providers serve it from $0.087 per million input tokens.
Verified 2026-10-25 Rank #79 in LLMs · score 103.2 method v1 Sources 5 Downloads 30d 139.4k · #280 of 619 tracked
Changed this week: 7 new quantized builds on the Hub (75 in all).2026-10-24 · 6 new quantized builds on the Hub (74 in all).2026-10-24 · 1 new provider: Nscale.2026-10-23
Picked by rule from 18 tracked pieces: the lab's own post, the most shared, a field report, the latest.
OpenAI released gpt-oss-120b and gpt-oss-20b, their first open-weight models since GPT-2 in 2019.
The 4 picks above, plus these 14. 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.
6 providers with a live price, from $0.087 to $0.85 per million input tokens; the median is $0.15. 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 (7 API providers); heavily run locally (75 quantized builds).
Top Spaces running it: hadadxyz/ai ♥ 819 · multimodalart/Qwen-Image-Edit-Fast ♥ 495 · AiSudo/Qwen-Image-to-LoRA ♥ 344 · burtenshaw/karpathy-llm-council ♥ 292
Weights only, an estimate: leave room for the KV cache. Method
About 141 GB at INT4, 282 GB at FP8, 564 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.
52.15. Each tick is a model on this site with a measured BFCL row; author-reported numbers stay in Reference. Percent (overall accuracy). Prompt
Source · leaderboard · 2026-10-25
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-qwen3-235b-a22b-instruct-2507",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 466,
"ttft_any_ms_median": 466,
"tokens_per_s_median": 40.1,
"runs": [
{
"ttft_s": 0.466,
"ttft_any_s": 0.466,
"tokens_per_s": 42.3,
"latency_s": 4.47,
"tokens_out": 189,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.454,
"ttft_any_s": 0.454,
"tokens_per_s": 38.7,
"latency_s": 4.85,
"tokens_out": 188,
"reasoning_chars": 0,
"finish": "stop"
},
{
"ttft_s": 0.507,
"ttft_any_s": 0.507,
"tokens_per_s": 40.1,
"latency_s": 4.37,
"tokens_out": 175,
"reasoning_chars": 0,
"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": "Your table at Cervejaria Ramiro has been successfully booked for Wednesday, 2026-10-14 at 19:30. Confirmation code: TBL-4471.",
"fixture": "receipts.errand.v1"
} {
"fields_matched": 80,
"fields_total": 80,
"cost_per_1000_usd": 0.0323,
"parsed": true,
"per_item": [
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
},
{
"street": true,
"city": true,
"postcode": true,
"country": true
}
],
"sample": [
{
"street": "1600 Amphitheatre Parkway",
"unit": null,
"city": "Mountain View",
"region": "CA",
"postcode": "94043",
"country": "US"
},
{
"street": "10 Downing Street",
"unit": null,
"city": "London",
"region": null,
"postcode": "SW1A 2AA",
"country": "GB"
},
{
"street": "Rua Augusta 274",
"unit": null,
"city": "Lisboa",
"region": null,
"postcode": "1100-053",
"country": "PT"
}
],
"fixture": "receipts.addresses.v1"
} Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-07-21), amber triangles are 3 dated articles (each links to it), grey ticks are repo commits.
104 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 BFCL · Call tools and APIs | 52.15 percent (overall accuracy) | Prompt5 sub-scores
| leaderboard | 2026-10-25 |
| measured BFCL · Call tools and APIs | 47.99 percent (overall accuracy) | FC5 sub-scores
| leaderboard | 2026-10-25 |
| author-reported AIME 2025 · Reason through hard problems | 70.3 percent | — | model author | 2026-10-07 |
| author-reported ARC-AGI · Reason through hard problems | 41.8 percent | — | model author | 2026-10-07 |
| author-reported Aider Polyglot · Write code with agents | 57.3 percent | — | model author | 2026-10-07 |
| author-reported Arena-Hard · Chat and follow instructions | 79.2 percent | — | model author | 2026-10-07 |
| author-reported GPQA Diamond · Reason through hard problems | 77.5 percent | — | model author | 2026-10-07 |
| author-reported HMMT · Reason through hard problems | 55.4 percent | — | model author | 2026-10-07 |
| author-reported IFEval · Chat and follow instructions | 88.7 percent | — | model author | 2026-10-07 |
| author-reported LiveBench · Reason through hard problems | 75.4 percent | — | model author | 2026-10-07 |
| author-reported LiveCodeBench · Write code with agents | 51.8 percent | — | model author | 2026-10-07 |
| author-reported MMLU · Reason through hard problems | 93.1 percent | — | model author | 2026-10-07 |
| author-reported MMLU-Pro · Reason through hard problems | 83 percent | — | model author | 2026-10-07 |
| author-reported MultiPL-E · Write code with agents | 87.9 percent | — | model author | 2026-10-07 |
| author-reported SimpleQA · Reason through hard problems | 54.3 percent | — | model author | 2026-10-07 |
| author-reported SimpleQA · Reason through hard problems | 84.3 percent | — | model author | 2026-10-07 |
| author-reported SuperGPQA · Reason through hard problems | 62.6 percent | — | model author | 2026-10-07 |
| author-reported ZebraLogic | 95 percent | — | 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 | Qwen3MoeForCausalLM |
| Model type | qwen3_moe |
| Dtype | bfloat16 |
| Params by dtype | BF16: 235,093,634,560 |
| Size band | XXL |
| MoE | yes |
| Experts | 128 |
| Experts per token | 8 |
| Context length | 262,144 (config.json max_position_embeddings) |
| Papers | arXiv:2402.17463, arXiv:2407.02490, arXiv:2501.15383, arXiv:2404.06654, arXiv:2505.09388 |
| License | Apache 2.0 (apache-2.0) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · Qwen3MoeForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · Qwen3MoeForCausalLM · 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 Qwen3-235B-A22B-Instruct-2507: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 564 GB · Multi-GPU node (<640 GB) |
|---|---|
| FP8 (1 byte/param) | 282 GB · Multi-GPU node (<640 GB) |
| INT4 (0.5 bytes/param) | 141 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: 75finetune: 20adapter: 9
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| Qwen/Qwen3-235B-A22B-Instruct-2507-FP8 | quantized | -FP8 | 235B | XXL | apache-2.0 | Qwen | 161 | 4.6M |
| lmstudio-community/Qwen3-235B-A22B-Instruct-2507-MLX-4bit | quantized | MLX | 235B | XXL | apache-2.0 | lmstudio-community | 1 | 277.1k |
| lmstudio-community/Qwen3-235B-A22B-Instruct-2507-MLX-8bit | quantized | MLX | 235B | XXL | apache-2.0 | lmstudio-community | 0 | 271.9k |
| lmstudio-community/Qwen3-235B-A22B-Instruct-2507-MLX-6bit | quantized | MLX | 235B | XXL | apache-2.0 | lmstudio-community | 0 | 270.8k |
| unsloth/Qwen3-235B-A22B-Instruct-2507-GGUF | quantized | GGUF | — | — | apache-2.0 | unsloth | 120 | 261k |
| QuantTrio/Qwen3-235B-A22B-Instruct-2507-AWQ | quantized | AWQ | 235B | XXL | apache-2.0 | QuantTrio | 13 | 250.4k |
| mlx-community/Qwen3-235B-A22B-Instruct-2507-4bit | quantized | -4bit | 235B | XXL | apache-2.0 | mlx-community | 3 | 105k |
| mlx-community/Qwen3-235B-A22B-Instruct-2507-8bit | quantized | -8bit | 235B | XXL | apache-2.0 | mlx-community | 2 | 94.9k |
| lmstudio-community/Qwen3-235B-A22B-Instruct-2507-GGUF | quantized | GGUF | — | — | — | lmstudio-community | 11 | 48.8k |
| nvidia/Qwen3-235B-A22B-Instruct-2507-NVFP4 | quantized | NVFP4 | 120B | XXL | apache-2.0 | nvidia | 13 | 45.2k |
| bartowski/Qwen_Qwen3-235B-A22B-Instruct-2507-GGUF | quantized | GGUF | — | — | — | bartowski | 9 | 33.9k |
| AIDXteam/Qwen3-235B-A22B-Instruct-2507-AWQ | quantized | AWQ | 32.3B | L | — | AIDXteam | 0 | 24.4k |
| RedHatAI/Qwen3-235B-A22B-Instruct-2507-NVFP4 | quantized | NVFP4 | 136B | XXL | apache-2.0 | RedHatAI | 4 | 21.2k |
| NVFP4/Qwen3-235B-A22B-Instruct-2507-FP4 | quantized | — | 118B | XXL | apache-2.0 | NVFP4 | 4 | 19.9k |
| amd/Qwen3-235B-A22B-Instruct-2507-MXFP4 | quantized | MXFP4 | 118B | XXL | apache-2.0 | amd | 2 | 14.7k |
| INC4AI/Qwen3-235B-A22B-Instruct-2507-gguf-q4km-AutoRound | quantized | gguf | — | — | apache-2.0 | INC4AI | 8 | 8.8k |
| ubergarm/Qwen3-235B-A22B-Instruct-2507-GGUF | quantized | GGUF | — | — | apache-2.0 | ubergarm | 17 | 8k |
| QuantTrio/Qwen3-235B-A22B-Instruct-2507-GPTQ-Int4-Int8Mix | quantized | GPTQ | 248B | XXL | apache-2.0 | QuantTrio | 4 | 6.1k |
| DevQuasar/Qwen.Qwen3-235B-A22B-Instruct-2507-GGUF | quantized | GGUF | — | — | — | DevQuasar | 2 | 4.9k |
| Intel/Qwen3-235B-A22B-Instruct-2507-int4-mixed-AutoRound | quantized | int4 | 1.29B | S | apache-2.0 | Intel | 12 | 3.5k |
| Intel/Qwen3-235B-A22B-Instruct-2507-gguf-q2ks-mixed-AutoRound | quantized | gguf | — | — | apache-2.0 | Intel | 15 | 3.2k |
| chriswritescode/Qwen3-235B-A22B-Instruct-2507-INT4-W4A16 | quantized | INT4 | 32.3B | L | apache-2.0 | chriswritescode | 0 | 3.9k |
| moxin-org/Qwen3-235B-A22B-Instruct-2507-GGUF | quantized | GGUF | — | — | apache-2.0 | moxin-org | 3 | 3.2k |
| unsloth/Qwen3-235B-A22B-Instruct-2507 | finetune | — | 235B | XXL | apache-2.0 | unsloth | 10 | 2.3k |
| John1604/Qwen3-235B-A22B-Instruct-2507-gguf | quantized | gguf | — | — | apache-2.0 | John1604 | 1 | 2.2k |
| manjunathshiva/Qwen3-235B-A22B-Instruct-2507-tq3a-tqTe-g64 | quantized | — | 16.4B | L | apache-2.0 | manjunathshiva | 1 | 1.8k |
| INCModel/Qwen3-235B-A22B-Instruct-2507-MXFP4-CT-AutoRound | quantized | MXFP4 | 8.00B | M | apache-2.0 | INCModel | 0 | 1.7k |
| sm54/Qwen3-235B-A22B-Instruct-2507-MXFP4_MOE | quantized | GGUF | — | — | — | sm54 | 9 | 1.1k |
| manjunathshiva/Qwen3-235B-A22B-Instruct-2507-tq3a-tq2e-g32 | quantized | — | 22.9B | L | apache-2.0 | manjunathshiva | 0 | 1.3k |
| nebius/EAGLE3-Qwen3-235B-A22B-Instruct-2507 | finetune | — | 573M | S | cc-by-4.0 | nebius | 0 | 1.3k |
| mlx-community/Qwen3-235B-A22B-Instruct-2507-3bit | quantized | — | 235B | XXL | apache-2.0 | mlx-community | 0 | 1.3k |
| wasmdashai/vivo-c-v1 | finetune | — | 235B | XXL | apache-2.0 | wasmdashai | 3 | 843 |
| jasonyux/Qwen3-235B-A22B-Instruct-2507-AWQ | quantized | AWQ | 235B | XXL | apache-2.0 | jasonyux | 0 | 861 |
| manjunathshiva/Qwen3-235B-A22B-Instruct-2507-tq3-g32 | quantized | — | 31.5B | L | apache-2.0 | manjunathshiva | 0 | 825 |
| derek-wh/Qwen3-235B-A22B-Instruct-2507-mlx-2Bit | quantized | mlx | 235B | XXL | apache-2.0 | derek-wh | 0 | 807 |
| mlx-community/Qwen3-235B-A22B-Instruct-2507-5bit | quantized | — | 235B | XXL | apache-2.0 | mlx-community | 1 | 678 |
| pcuenq/Qwen3-235B-A22B-Instruct-2507-mlx-4Bit | quantized | mlx | 235B | XXL | apache-2.0 | pcuenq | 1 | 653 |
| zjiayu064/Qwen3-235B-A22B-Instruct-2507-BitsMoE-1.6bit | quantized | — | 21.5B | L | apache-2.0 | zjiayu064 | 0 | 676 |
| bobchenyx/Qwen3-235B-A22B-Instruct-2507-GGUF | quantized | GGUF | — | — | mit | bobchenyx | 1 | 602 |
| locailabs/locai-l1-large | finetune | — | 235B | XXL | apache-2.0 | locailabs | 8 | 211 |
| Date | Commit | SHA |
|---|---|---|
| 2025-09-17 | Update tokenizer_config.json | ac9c66c |
| 2025-08-17 | Update README.md | 97d5e9a |
| 2025-08-08 | Add 1M support (#30) | 8994a5c |
| 2025-07-30 | Update README.md | 9fe72e5 |
| 2025-07-29 | Update README.md | 8b0a014 |
| 2025-07-21 | Update README.md | 56e16a6 |
| 2025-07-21 | Update README.md | 61e22c6 |
| 2025-07-21 | Update README.md | 8a39f5b |
| 2025-07-21 | Update README.md | bbd8c67 |
| 2025-07-21 | Update README.md | 8399681 |
| 2025-07-21 | Create LICENSE | 585555b |
| 2025-07-21 | Update tokenizer_config.json | 3d2db25 |
| 2025-07-21 | Update README.md | 0f2e333 |
| 2025-07-21 | Delete configuration.json | a46ff03 |
| 2025-07-21 | Delete .mv | eea9174 |
| 2025-07-21 | Delete .msc | 4ee6ddd |
| 2025-07-21 | Add files using upload-large-folder tool | d4dc8b0 |
| 2025-07-21 | initial commit | a1c9a1b |
Hosted endpoints observed for Qwen3-235B-A22B-Instruct-2507, 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: $0.09/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| GMICloud openrouter | qwen/qwen3-235b-a22b-2507 | 256K | fp8 | $0.09 | $0.35 | — | — | yes | live | 2026-10-24 |
| DeepInfra hf, deepinfra, openrouter | Qwen/Qwen3-235B-A22B-Instruct-2507 | 256K | fp8 | $0.09 | $0.55 | 35 | 200 ms | yes | live | 2026-10-25 |
| Parasail parasail, openrouter | parasail-qwen3-235b-a22b-instruct-2507 | 128K | fp8 | $0.14 | $0.80 | — | — | yes | serverless | 2026-10-24 |
| Venice openrouter | qwen/qwen3-235b-a22b-2507 | 125K | fp8 | $0.15 | $0.75 | — | — | yes | live | 2026-10-24 |
| Nscale hf | Qwen/Qwen3-235B-A22B-Instruct-2507 | 32K | — | $0.20 | $0.60 | 16.3 | 433 ms | yes | live | 2026-10-25 |
| Google openrouter | qwen/qwen3-235b-a22b-2507 | 256K | unknown | $0.22 | $0.88 | — | — | yes | status -5 | 2026-10-24 |
| Scaleway hf | qwen3-235b-a22b-instruct-2507 | — | — | $0.85 | $2.56 | 30.4 | 1011 ms | yes | 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 21:32 UTC · Providers: 2026-10-24 21:44 UTC · Articles: 2026-10-24 21:45 UTC · Summary: 2026-10-07 15:00 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.