Up to date
deepseek-ai MIT out 11 months ago
685B parameters (8 of 256 experts per token), text in, text out, 160K tokens of context. 11 providers serve it from $0.26 per million input tokens.
Verified 2026-10-25 Rank #65 in LLMs · score 104.7 method v1 Sources 4 Downloads 30d 5.6M · #26 of 619 tracked
Changed this week: 1 new quantized build on the Hub (20 in all).2026-10-24
Picked by rule from 28 tracked pieces: the lab's own post, the most shared, a field report, the latest.
DeepSeek launches V3.2 and V3.2-Speciale, reasoning-first models built for agents with V3.2-Speciale attaining gold-level results in IMO, CMO, ICPC World Finals & IOI 2025.
DeepSeek-V3.2-Exp introduces DeepSeek Sparse Attention to optimize long-context efficiency while maintaining performance on par with V3.1-Terminus.
DeepSeek-V3.2 is an open-weights language model for text generation published by deepseek-ai and added to Hugging Bay Jun 29, 2026.
The 3 picks above, plus these 22. 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.
9 providers with a live price, from $0.26 to $3.00 per million input tokens; the median is $0.28. 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 (11 API providers); 20 quantized builds for local use.
Top Spaces running it: enzostvs/deepsite ♥ 16,604 · hadadxyz/ai ♥ 819 · embedl/hfviewer ♥ 181 · hfmlsoc/different-flops ♥ 11
Weights only, an estimate: leave room for the KV cache. Method
About 411 GB at INT4, 822 GB at FP8, 1.64 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
Third-party measurements only; author-reported numbers stay in Reference, marked as such.
70. Each tick is a model on this site with a measured SWE-bench Verified row; author-reported numbers stay in Reference. Percent resolved. agent: mini-SWE-agent, reasoning_effort=high
Source · leaderboard · 2026-02-17
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-12-01), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.
62 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 Multilingual · Write code with agents | 59 percent resolved | agent: mini-SWE-agent | leaderboard | 2026-02-13 |
| measured SWE-bench Verified · Write code with agents | 70 percent resolved | agent: mini-SWE-agent, reasoning_effort=high | leaderboard | 2026-02-17 |
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 | DeepseekV32ForCausalLM |
| Model type | deepseek_v32 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 3,946,184,704 · F32: 1,075,456 · F8_E4M3: 681,408,069,632 |
| Size band | XXL |
| MoE | yes |
| Experts | 256 |
| Experts per token | 8 |
| Context length | 163,840 (config.json max_position_embeddings) |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · DeepseekV32ForCausalLM · 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 DeepSeek-V3.2: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 1.64 TB · Cluster (≥640 GB) |
|---|---|
| FP8 (1 byte/param) | 822 GB · Cluster (≥640 GB) |
| INT4 (0.5 bytes/param) | 411 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.
Base model: deepseek-ai/DeepSeek-V3.2-Exp-Base
quantized: 20finetune: 34adapter: 7merge: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| QuantTrio/DeepSeek-V3.2-AWQ | quantized | AWQ | 685B | XXL | mit | QuantTrio | 11 | 1.5M |
| nvidia/DeepSeek-V3.2-NVFP4 | finetune | NVFP4 | 394B | XXL | mit | nvidia | 19 | 175.3k |
| DevQuasar/deepseek-ai.DeepSeek-V3.2-GGUF | quantized | GGUF | — | — | — | DevQuasar | 1 | 161.5k |
| unsloth/DeepSeek-V3.2-GGUF | quantized | GGUF | — | — | mit | unsloth | 41 | 101.1k |
| mlx-community/DeepSeek-V3.2-4bit | finetune | -4bit | 672B | XXL | mit | mlx-community | 7 | 18.4k |
| amd/DeepSeek-V3.2-mxfp4 | quantized | mxfp4 | 377B | XXL | mit | amd | 1 | 9.8k |
| cerebras/DeepSeek-V3.2-REAP-345B-A37B | quantized | — | 345B | XXL | mit | cerebras | 34 | 7.4k |
| lovedheart/DeepSeek-V3.2-REAP-345B-A37B-GGUF-Experimental | quantized | GGUF | — | — | mit | lovedheart | 4 | 6.1k |
| mlx-community/DeepSeek-V3.2-8bit | finetune | -8bit | 672B | XXL | mit | mlx-community | 0 | 4.8k |
| mlx-community/DeepSeek-V3.2-mlx-5bit | quantized | mlx | 672B | XXL | mit | mlx-community | 2 | 4.3k |
| bdellabe/DeepSeek-V3.2-NVFP4-FP8-BLOCK | quantized | NVFP4 | 384B | XXL | apache-2.0 | bdellabe | 0 | 3.8k |
| cerebras/DeepSeek-V3.2-REAP-508B-A37B | quantized | — | 508B | XXL | mit | cerebras | 15 | 3k |
| mlx-community/DeepSeek-V3.2_bf16 | finetune | — | 672B | XXL | mit | mlx-community | 1 | 3.4k |
| baa-ai/DeepSeek-V3.2-RAM-350GB-MLX | quantized | MLX | 672B | XXL | mit | baa-ai | 0 | 3k |
| sjug/DeepSeek-V3.2-MLX-4bit | finetune | MLX | 672B | XXL | mit | sjug | 0 | 2.7k |
| exolabs/DeepSeek-V3.2_bf16 | finetune | — | 672B | XXL | mit | exolabs | 0 | 2.5k |
| RedHatAI/DeepSeek-V3.2-NVFP4-FP8-BLOCK | quantized | NVFP4 | 384B | XXL | mit | RedHatAI | 1 | 2.2k |
| amd/DeepSeek-V3.2-mtp-ptpc | quantized | — | 686B | XXL | mit | amd | 1 | 1.7k |
| sszymczyk/DeepSeek-V3.2-nolight-GGUF | quantized | GGUF | — | — | mit | sszymczyk | 10 | 1.2k |
| ubicloud/DeepSeek-V3.2-Pruned-15B | quantized | — | 15.2B | L | — | ubicloud | 1 | 624 |
| sarthak2314/indicbart-hindi-gec-v2 | finetune | — | 244M | S | — | sarthak2314 | 0 | 629 |
| Doctor-Shotgun/DeepSeek-V3.2-dense-attn-GGUF | quantized | GGUF | — | — | — | Doctor-Shotgun | 3 | 278 |
| lovedheart/DeepSeek-V3.2-GGUF-Experimental | quantized | GGUF | — | — | mit | lovedheart | 1 | 292 |
| kathywu95/deepseek-v3.2-small-random-fp8-64-experts | quantized | -fp8 | 5.52B | M | — | kathywu95 | 1 | 280 |
| CPU-Hybrid-MoE/DeepSeek-V3.2-CPU-NUMA4-AMXINT8 | quantized | — | — | — | — | CPU-Hybrid-MoE | 1 | 150 |
| Kebob/DeepSeek-V3.2-CPU-NUMA2-AMXINT4 | quantized | — | — | — | mit | Kebob | 0 | 172 |
| ckoh04/DeepSeek-V3.2-NVFP4-active-slice | finetune | NVFP4 | 25.2B | L | mit | ckoh04 | 0 | 129 |
| Hotblaz/Compare_Anti_9 | finetune | — | — | — | mit | Hotblaz | 1 | 2 |
| Dominic586/ugly | finetune | — | — | — | apache-2.0 | Dominic586 | 1 | 0 |
| nvcnvn/DeepSeek-V3.2-4L | quantized | — | 28.6B | L | mit | nvcnvn | 0 | 19 |
| Syntrallm/syntrallm | adapter | — | — | — | mit | Syntrallm | 0 | 0 |
| massmas/ok | finetune | — | — | — | — | massmas | 0 | 0 |
| KNB86/ccr | finetune | — | — | — | — | KNB86 | 0 | 0 |
| cridhdxbgct5/bb | adapter | — | — | — | apache-2.0 | cridhdxbgct5 | 0 | 0 |
| Yooghrn/Goodheng | adapter | — | — | — | bigcode-openrail-m | Yooghrn | 0 | 0 |
| Adimasdn/Humaniora | finetune | — | — | — | apache-2.0 | Adimasdn | 0 | 0 |
| mengerti/my-robot-dataset | finetune | — | — | — | apache-2.0 | mengerti | 0 | 0 |
| Maxmaxjawa/max1412 | finetune | — | — | — | llama4 | Maxmaxjawa | 0 | 0 |
| wedangjahe/Greenday | finetune | — | — | — | mit | wedangjahe | 0 | 0 |
| Arief1899X/Galuh | finetune | — | — | — | apache-2.0 | Arief1899X | 0 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2025-12-01 | Update README.md | a7e62ac |
| 2025-12-01 | Release DeepSeek-V3.2 | c69397e |
| 2025-12-01 | Add files using upload-large-folder tool | 205ad63 |
| 2025-12-01 | Add files using upload-large-folder tool | 00846f3 |
| 2025-12-01 | Add files using upload-large-folder tool | fba8af1 |
| 2025-12-01 | Add files using upload-large-folder tool | ba658fb |
| 2025-12-01 | initial commit | e91970e |
Hosted endpoints observed for DeepSeek-V3.2, 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: $0.21/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| GMICloud openrouter | deepseek/deepseek-v3.2 | 160K | fp8 | $0.21 | $0.31 | — | — | yes | status -2 | 2026-10-24 |
| SiliconFlow openrouter | deepseek/deepseek-v3.2 | 160K | fp8 | $0.26 | $0.80 | — | — | yes | live | 2026-10-24 |
| DeepInfra hf, deepinfra, openrouter | deepseek-ai/DeepSeek-V3.2 | 16K | fp4 | $0.26 | $0.38 | 32.5 | 440 ms | yes | live | 2026-10-25 |
| AtlasCloud openrouter | deepseek/deepseek-v3.2 | 160K | fp8 | $0.26 | $0.38 | — | — | yes | live | 2026-10-24 |
| Venice openrouter | deepseek/deepseek-v3.2 | 156K | unknown | $0.27 | $0.39 | — | — | yes | live | 2026-10-24 |
| Baidu openrouter | deepseek/deepseek-v3.2 | 128K | fp8 | $0.28 | $0.42 | — | — | — | live | 2026-10-24 |
| DigitalOcean openrouter | deepseek/deepseek-v3.2 | 160K | unknown | $0.30 | $0.96 | — | — | — | live | 2026-10-24 |
| Google openrouter | deepseek/deepseek-v3.2 | 160K | unknown | $0.56 | $1.68 | — | — | yes | status -5 | 2026-10-24 |
| Phala openrouter | deepseek/deepseek-v3.2 | 160K | unknown | $1.00 | $1.00 | — | — | yes | live | 2026-10-24 |
| Mara openrouter | deepseek/deepseek-v3.2 | 32K | unknown | $3.00 | $4.50 | — | — | — | live | 2026-10-24 |
| SambaNova openrouter | deepseek/deepseek-v3.2 | 32K | unknown | $3.00 | $4.50 | — | — | — | 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.
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.
