# DeepSeek-V3 by DeepSeek: Open-Weight Model Source: https://savrn.com/models/deepseek-v3 Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve DeepSeek-V3 (684.5B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them. | Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits | | --- | --- | --- | --- | --- | --- | | 16-bit | 1369.0 GB | 1642.8 GB | 7x [MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) (256 GB) Vultr | $14.00 | [6x MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) $15.54 · [7x B300](https://savrn.com/ai-index/pricing/gpus/b300) $46.20 | | 8-bit | 684.5 GB | 821.4 GB | 3x [MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) (288 GB) Vultr | $7.77 | [4x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $8.00 · [5x MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) $9.25 | | 4-bit | 342.2 GB | 410.7 GB | 2x [MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) (256 GB) Vultr | $4.00 | [2x MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) $5.18 · [3x MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) $5.55 | Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the [SAVRN Index](https://savrn.com/ai-index/pricing/gpus), read Oct 24, 2026. [DeepSeek-V3 on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/deepseek-v3/gpus) ## SAVRN's Notes on DeepSeek-V3 1,642.9 GB. That is what DeepSeek-V3 needs at 16-bit, so the cheapest Index setup there is seven MI325X cards at $14.00 per hour. At 8-bit it drops to 821.4 GB on three MI355X at $7.77; at 4-bit, 410.7 GB on two MI325X at $4.00. What makes a 684.5B-parameter text model livable is the routing: 256 experts, 8 active per token, so compute per token is a fraction of the memory you provision. The 163,840-token context takes whole document sets in one call. Our facts carry no license for this model, so get the license text from the publisher before committing hardware. The weights are 185 files and 688.6 GB, so plan transfer and storage ahead of the cards. And run the math against the API: DeepInfra lists $0.32 per million input tokens and $0.89 output, so a $4.00-per-hour 4-bit rig needs a known token volume to pay off. SAVRN Research, 2026-09-18 ## Model Card We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations… Excerpt from the card by DeepSeek. ## Configuration Architecture DeepseekV3ForCausalLM Context length (tokens) 163,840 Layers 61 Hidden size 7,168 Feed-forward size 18,432 Attention heads 128 Key/value heads 128 Vocabulary size 129,280 Routed experts 256 Experts active per token 8 RoPE base 10,000 Stored precision bfloat16 Model type deepseek_v3 Quantization fp8 ## Identity and Version Repository deepseek-ai/DeepSeek-V3 Publisher DeepSeek Task Text generation Modality Text Library transformers Parameters 684.5B parameters Languages Not stated by the source Revision e815299b0bcbac849fa540c768ef21845365c9eb First published 2024-12-25 Last updated 2025-03-27 ## Files and Weights 185 files, 688.6 GB in total. The weights are 163 files totalling 688.6 GB in safetensors. Weights163 files · 688.6 GB Configuration12 files · 9.0 MB Tokenizer2 files · 7.9 MB Documentation4 files · 38.1 KB Other3 files · 292.1 KB Repository1 file · 1.5 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | model-00001-of-000163.safetensors | Weights | 5.2 GB | b933b099f335 | | model-00002-of-000163.safetensors | Weights | 4.3 GB | f619dca0db00 | | model-00003-of-000163.safetensors | Weights | 4.3 GB | 1f23a8b42d45 | | model-00004-of-000163.safetensors | Weights | 4.3 GB | 423e7a58cbb9 | | model-00005-of-000163.safetensors | Weights | 4.3 GB | 8613aed68453 | | model-00006-of-000163.safetensors | Weights | 4.4 GB | 7938cb303f2c | | model-00007-of-000163.safetensors | Weights | 4.3 GB | 3e4d025d5510 | | model-00008-of-000163.safetensors | Weights | 4.3 GB | a1c86b74e9cb | | model-00009-of-000163.safetensors | Weights | 4.3 GB | 0a9a0804e3fb | | model-00010-of-000163.safetensors | Weights | 4.3 GB | ba45df8c31e8 | | model-00011-of-000163.safetensors | Weights | 4.3 GB | 268851646260 | | model-00012-of-000163.safetensors | Weights | 1.3 GB | 71986c26dcc3 | | model-00013-of-000163.safetensors | Weights | 4.3 GB | befea8fe1acb | | model-00014-of-000163.safetensors | Weights | 4.3 GB | 53a9be88a0aa | | model-00015-of-000163.safetensors | Weights | 4.3 GB | a1220514c2f0 | | model-00016-of-000163.safetensors | Weights | 4.3 GB | 0838ed15d5bb | | model-00017-of-000163.safetensors | Weights | 4.3 GB | c56501d42f3e | | model-00018-of-000163.safetensors | Weights | 4.3 GB | 15ba8e1a5149 | | model-00019-of-000163.safetensors | Weights | 4.3 GB | 1e21f56c613a | | model-00020-of-000163.safetensors | Weights | 4.3 GB | 56d73a3442aa | | model-00021-of-000163.safetensors | Weights | 4.3 GB | 3465da1ad357 | | model-00022-of-000163.safetensors | Weights | 4.3 GB | 00a47040de71 | | model-00023-of-000163.safetensors | Weights | 4.3 GB | 2b1972709df3 | | model-00024-of-000163.safetensors | Weights | 4.3 GB | d86d6a40bc05 | | model-00025-of-000163.safetensors | Weights | 4.3 GB | 888c83fba1bb | | model-00026-of-000163.safetensors | Weights | 4.3 GB | 568555db6a0a | | model-00027-of-000163.safetensors | Weights | 4.3 GB | 98656f5ecfe6 | | model-00028-of-000163.safetensors | Weights | 4.3 GB | 1c45e4a3ba2c | | model-00029-of-000163.safetensors | Weights | 4.3 GB | 286a4e7f47c5 | | model-00030-of-000163.safetensors | Weights | 4.3 GB | c11ead3c5f4a | | model-00031-of-000163.safetensors | Weights | 4.3 GB | e94d32e8649e | | model-00032-of-000163.safetensors | Weights | 4.3 GB | e7333388eedf | | model-00033-of-000163.safetensors | Weights | 4.3 GB | f6b8eded7cb9 | | model-00034-of-000163.safetensors | Weights | 1.7 GB | bdf369e4ecb2 | | model-00035-of-000163.safetensors | Weights | 4.3 GB | 5c7afe2627b1 | | model-00036-of-000163.safetensors | Weights | 4.3 GB | 2eb3e5f66889 | | model-00037-of-000163.safetensors | Weights | 4.3 GB | f5eedc7c4581 | | model-00038-of-000163.safetensors | Weights | 4.3 GB | 5be6811ab78a | | model-00039-of-000163.safetensors | Weights | 4.3 GB | be4ce145bc0a | | model-00040-of-000163.safetensors | Weights | 4.3 GB | b421975c36f7 | | model-00041-of-000163.safetensors | Weights | 4.3 GB | 98cb15c98687 | | model-00042-of-000163.safetensors | Weights | 4.3 GB | 69d2383fe605 | | model-00043-of-000163.safetensors | Weights | 4.3 GB | f5a9e6a958dc | | model-00044-of-000163.safetensors | Weights | 4.3 GB | 2d32eff0a6ce | | model-00045-of-000163.safetensors | Weights | 4.3 GB | 1b5aeae7ecce | | model-00046-of-000163.safetensors | Weights | 4.3 GB | f842c6a79d05 | | model-00047-of-000163.safetensors | Weights | 4.3 GB | e1ab502eae5b | | model-00048-of-000163.safetensors | Weights | 4.3 GB | 7cabe3dc1d4b | | model-00049-of-000163.safetensors | Weights | 4.3 GB | 0af950690c49 | | model-00050-of-000163.safetensors | Weights | 4.3 GB | 805ca3bc43e3 | | model-00051-of-000163.safetensors | Weights | 4.3 GB | 76e249f3e1e8 | | model-00052-of-000163.safetensors | Weights | 4.3 GB | ba4543f76284 | | model-00053-of-000163.safetensors | Weights | 4.3 GB | c726a7c86b4f | | model-00054-of-000163.safetensors | Weights | 4.3 GB | 09bf2a5fc1cc | | model-00055-of-000163.safetensors | Weights | 4.3 GB | 12ea0f770efa | | model-00056-of-000163.safetensors | Weights | 1.7 GB | 1e094059b3f5 | | model-00057-of-000163.safetensors | Weights | 4.3 GB | 86f30117b460 | | model-00058-of-000163.safetensors | Weights | 4.3 GB | b131e54bc6bd | | model-00059-of-000163.safetensors | Weights | 4.3 GB | c6cf937071dd | | model-00060-of-000163.safetensors | Weights | 4.3 GB | d04f96624640 | | model-00061-of-000163.safetensors | Weights | 4.3 GB | 122af5c0a0ca | | model-00062-of-000163.safetensors | Weights | 4.3 GB | 3845f9eba351 | | model-00063-of-000163.safetensors | Weights | 4.3 GB | 8bc190d0f67b | | model-00064-of-000163.safetensors | Weights | 4.3 GB | f8d8d94619a1 | | model-00065-of-000163.safetensors | Weights | 4.3 GB | f0bda27a01c5 | | model-00066-of-000163.safetensors | Weights | 4.3 GB | 4f907d245b6d | | model-00067-of-000163.safetensors | Weights | 4.3 GB | 5107a41e1771 | | model-00068-of-000163.safetensors | Weights | 4.3 GB | fa5187cd0f09 | | model-00069-of-000163.safetensors | Weights | 4.3 GB | 31707005180b | | model-00070-of-000163.safetensors | Weights | 4.3 GB | a350ac1e4754 | | model-00071-of-000163.safetensors | Weights | 4.3 GB | 60424a0f4c25 | | model-00072-of-000163.safetensors | Weights | 4.3 GB | 6b0b6f9a1fb3 | | model-00073-of-000163.safetensors | Weights | 4.3 GB | 4d386236d747 | | model-00074-of-000163.safetensors | Weights | 4.3 GB | f663d4e39baa | | model-00075-of-000163.safetensors | Weights | 4.3 GB | 712e24abbb57 | | model-00076-of-000163.safetensors | Weights | 4.3 GB | 5f922ddc4f32 | | model-00077-of-000163.safetensors | Weights | 4.3 GB | c299d2121231 | | model-00078-of-000163.safetensors | Weights | 1.7 GB | 7fa2a67820d2 | | model-00079-of-000163.safetensors | Weights | 4.3 GB | 38fb787ec3ed | | model-00080-of-000163.safetensors | Weights | 4.3 GB | 07556b85b207 | | model-00081-of-000163.safetensors | Weights | 4.3 GB | 92c6b2c837a4 | | model-00082-of-000163.safetensors | Weights | 4.3 GB | 28d9cc27f97d | | model-00083-of-000163.safetensors | Weights | 4.3 GB | 9390263503c9 | | model-00084-of-000163.safetensors | Weights | 4.3 GB | c254740ff8e4 | | model-00085-of-000163.safetensors | Weights | 4.3 GB | b832c3b79a8a | | model-00086-of-000163.safetensors | Weights | 4.3 GB | 7cf824bf62e6 | | model-00087-of-000163.safetensors | Weights | 4.3 GB | af791e4477bd | | model-00088-of-000163.safetensors | Weights | 4.3 GB | 6f7db096b2cc | | model-00089-of-000163.safetensors | Weights | 4.3 GB | 2129867f49a1 | | model-00090-of-000163.safetensors | Weights | 4.3 GB | 7fab650d1ed4 | | model-00091-of-000163.safetensors | Weights | 4.3 GB | 519a413487c5 | | model-00092-of-000163.safetensors | Weights | 4.3 GB | e830585cb310 | | model-00093-of-000163.safetensors | Weights | 4.3 GB | ea2b9fc654be | | model-00094-of-000163.safetensors | Weights | 4.3 GB | c44b1d39f801 | | model-00095-of-000163.safetensors | Weights | 4.3 GB | fe8cf8717b94 | | model-00096-of-000163.safetensors | Weights | 4.3 GB | 18af8ddf19f2 | | model-00097-of-000163.safetensors | Weights | 4.3 GB | 3b6b7f63017d | | model-00098-of-000163.safetensors | Weights | 4.3 GB | 69825814a5ae | | model-00099-of-000163.safetensors | Weights | 4.3 GB | 495dd6d48845 | | model-00100-of-000163.safetensors | Weights | 1.7 GB | 8e14037bf546 | | model-00101-of-000163.safetensors | Weights | 4.3 GB | d2e75ba894c2 | | model-00102-of-000163.safetensors | Weights | 4.3 GB | c52f51a3e20e | | model-00103-of-000163.safetensors | Weights | 4.3 GB | 86652a9d272c | | model-00104-of-000163.safetensors | Weights | 4.3 GB | ecde06a98c84 | | model-00105-of-000163.safetensors | Weights | 4.3 GB | dd3836e12e35 | | model-00106-of-000163.safetensors | Weights | 4.3 GB | a4c63a600528 | | model-00107-of-000163.safetensors | Weights | 4.3 GB | a4e1a19f5340 | | model-00108-of-000163.safetensors | Weights | 4.3 GB | 3e1efd9300e6 | | model-00109-of-000163.safetensors | Weights | 4.3 GB | fca398562437 | | model-00110-of-000163.safetensors | Weights | 4.3 GB | dac3c17b6f31 | | model-00111-of-000163.safetensors | Weights | 4.3 GB | e8d0b963898c | | model-00112-of-000163.safetensors | Weights | 4.3 GB | 0c7030edeb4d | | model-00113-of-000163.safetensors | Weights | 4.3 GB | 2ca5c1ad7b1d | | model-00114-of-000163.safetensors | Weights | 4.3 GB | f7bc372c8e65 | | model-00115-of-000163.safetensors | Weights | 4.3 GB | 336a2969eb6b | | model-00116-of-000163.safetensors | Weights | 4.3 GB | 1cb0cb2c7437 | | model-00117-of-000163.safetensors | Weights | 4.3 GB | d160873806f7 | | model-00118-of-000163.safetensors | Weights | 4.3 GB | 72680742383e | | model-00119-of-000163.safetensors | Weights | 4.3 GB | a7f04447b66d | | model-00120-of-000163.safetensors | Weights | 4.3 GB | a540ce2f0766 | | model-00121-of-000163.safetensors | Weights | 4.3 GB | e37eecb8bf6e | | model-00122-of-000163.safetensors | Weights | 1.7 GB | 1c5691a0293c | | model-00123-of-000163.safetensors | Weights | 4.3 GB | c7f647d98cb4 | | model-00124-of-000163.safetensors | Weights | 4.3 GB | f3ddfe99d4b4 | | model-00125-of-000163.safetensors | Weights | 4.3 GB | 0558a282fb72 | | model-00126-of-000163.safetensors | Weights | 4.3 GB | 79744a5d57eb | | model-00127-of-000163.safetensors | Weights | 4.3 GB | bc29d0be4287 | | model-00128-of-000163.safetensors | Weights | 4.3 GB | fe5b46ca8a8a | | model-00129-of-000163.safetensors | Weights | 4.3 GB | 5ccf311da30b | | model-00130-of-000163.safetensors | Weights | 4.3 GB | 1ef6f7e7e0fa | | model-00131-of-000163.safetensors | Weights | 4.3 GB | 13265becdc7a | | model-00132-of-000163.safetensors | Weights | 4.3 GB | 66f55bd2bb59 | | model-00133-of-000163.safetensors | Weights | 4.3 GB | 2a810f80e3a1 | | model-00134-of-000163.safetensors | Weights | 4.3 GB | 710bd36d7f3a | | model-00135-of-000163.safetensors | Weights | 4.3 GB | dfa8902cbe1f | | model-00136-of-000163.safetensors | Weights | 4.3 GB | 95771ee9e65c | | model-00137-of-000163.safetensors | Weights | 4.3 GB | e5099a159fa7 | | model-00138-of-000163.safetensors | Weights | 4.3 GB | ce57c5d6ca9f | | model-00139-of-000163.safetensors | Weights | 4.3 GB | e45e57d8b348 | | model-00140-of-000163.safetensors | Weights | 4.3 GB | 64e8d879017d | | model-00141-of-000163.safetensors | Weights | 3.1 GB | 53a063da2194 | | model-00142-of-000163.safetensors | Weights | 4.3 GB | 86161cab1bba | | model-00143-of-000163.safetensors | Weights | 4.3 GB | c154533d5dd9 | | model-00144-of-000163.safetensors | Weights | 4.3 GB | 9cfe71f25918 | | model-00145-of-000163.safetensors | Weights | 4.3 GB | 6e75678f19b4 | | model-00146-of-000163.safetensors | Weights | 4.3 GB | a73fa1faff5d | | model-00147-of-000163.safetensors | Weights | 4.3 GB | 7b787dd38a96 | | model-00148-of-000163.safetensors | Weights | 4.3 GB | 229518f00be3 | | model-00149-of-000163.safetensors | Weights | 4.3 GB | e87c53c353c3 | | model-00150-of-000163.safetensors | Weights | 4.3 GB | 24f203609ab4 | | model-00151-of-000163.safetensors | Weights | 4.3 GB | 68909c3f465d | | model-00152-of-000163.safetensors | Weights | 4.3 GB | e1a3478e9d55 | | model-00153-of-000163.safetensors | Weights | 4.3 GB | 7df9a5d1f1e3 | | model-00154-of-000163.safetensors | Weights | 4.3 GB | 12a3c02867c9 | | model-00155-of-000163.safetensors | Weights | 4.3 GB | db7df665d353 | | model-00156-of-000163.safetensors | Weights | 4.3 GB | 88c75db2a58a | | model-00157-of-000163.safetensors | Weights | 4.3 GB | 1da4a1f20375 | | model-00158-of-000163.safetensors | Weights | 4.3 GB | 9c12b470327e | | model-00159-of-000163.safetensors | Weights | 4.3 GB | 5e4be3f94f2d | | model-00160-of-000163.safetensors | Weights | 5.2 GB | d99c3c4ebe91 | | model-00161-of-000163.safetensors | Weights | 4.3 GB | ccaa50981675 | | model-00162-of-000163.safetensors | Weights | 4.3 GB | b233f58b3441 | | model-00163-of-000163.safetensors | Weights | 6.6 GB | 1ad67ea80cad | | config.json | Configuration | 1.7 KB | — | | configuration_deepseek.py | Configuration | 9.9 KB | — | | inference/configs/config_16B.json | Configuration | 417 B | — | | inference/configs/config_236B.json | Configuration | 455 B | — | | inference/configs/config_671B.json | Configuration | 503 B | — | | inference/convert.py | Configuration | 3.2 KB | — | | inference/fp8_cast_bf16.py | Configuration | 3.2 KB | — | | inference/generate.py | Configuration | 5.4 KB | — | | inference/kernel.py | Configuration | 4.3 KB | — | | inference/model.py | Configuration | 17.6 KB | — | | model.safetensors.index.json | Configuration | 8.9 MB | — | | modeling_deepseek.py | Configuration | 75.7 KB | — | | LICENSE-CODE | Documentation | 1.1 KB | — | | LICENSE-MODEL | Documentation | 13.8 KB | — | | README.md | Documentation | 19.6 KB | — | | README_WEIGHTS.md | Documentation | 3.6 KB | — | | figures/benchmark.png | Other | 183.6 KB | — | | figures/niah.png | Other | 108.5 KB | — | | inference/requirements.txt | Other | 66 B | — | | .gitattributes | Repository | 1.5 KB | — | | tokenizer.json | Tokenizer | 7.8 MB | — | | tokenizer_config.json | Tokenizer | 3.1 KB | — | ## License and Download License Not stated by the source Access Open weights, no gate Download size 688.6 GB [Download from DeepSeek](https://huggingface.co/deepseek-ai/DeepSeek-V3) Released by DeepSeek through its official repository on Hugging Face. ## Built From - Described by [arXiv:2412.19437](https://savrn.com/papers/deepseek-v3-technical-report) ## Evaluations Each result is shown as reported, with the conditions its reporter stated. None is a SAVRN measurement. A comparison lines two results up only when their configuration, unit and setup are all stated and identical. | Benchmark | Conditions | Result | Reported by | Revision | Date | | --- | --- | --- | --- | --- | --- | | Idavidrein/gpqa | Task diamondMetric diamondSetup GPQA DiamondComparison conditions not established | 58.2071 | [EvalEval](https://huggingface.co/datasets/evaleval/EEE_datastore/blob/192329fb7d6b15b7b0936a1a58ae862aa7e8ba24/flat/objects/0a/db/0adbcd9f-38e6-414a-88b1-7bd928d4c72e.json) Reported by a third party | Evaluated revision not stated | 2025-03-19 | | LEXam-Benchmark/LEXam | Task mcq_4_choicesMetric mcq_4_choicesComparison conditions not established | 46.57 | LEXam Leaderboard Reported by a third party | Evaluated revision not stated | 2026-06-02 | | LEXam-Benchmark/LEXam | Task open_questionMetric open_questionComparison conditions not established | 52.53 | LEXam Leaderboard Reported by a third party | Evaluated revision not stated | 2026-06-02 | | TIGER-Lab/MMLU-Pro | Task mmlu_proMetric mmlu_proComparison conditions not established | 64.4 | [Model Card](https://huggingface.co/deepseek-ai/DeepSeek-V3) Reported by a third party | Evaluated revision not stated | 2026-01-28 | | openai/gsm8k | Task gsm8kMetric gsm8kComparison conditions not established | 89.3 | [Model Card](https://huggingface.co/deepseek-ai/DeepSeek-V3) Reported by a third party | Evaluated revision not stated | 2024-12-25 | | thamilvendhan/signalbench | Task access_denyMetric access_denySetup family=access_deny; n=12Comparison conditions not established | 0.5833 | [thamilvendhan](https://huggingface.co/datasets/thamilvendhan/signalbench) Reported by a third party | Evaluated revision not stated | 2026-07-08 | | thamilvendhan/signalbench | Task bot_policyMetric bot_policySetup family=bot_policy; n=12Comparison conditions not established | 0.25 | [thamilvendhan](https://huggingface.co/datasets/thamilvendhan/signalbench) Reported by a third party | Evaluated revision not stated | 2026-07-08 | | thamilvendhan/signalbench | Task injectionMetric injectionSetup family=injection; n=12Comparison conditions not established | 0.5833 | [thamilvendhan](https://huggingface.co/datasets/thamilvendhan/signalbench) Reported by a third party | Evaluated revision not stated | 2026-07-08 | | thamilvendhan/signalbench | Task memory_labelMetric memory_labelSetup family=memory_label; n=12Comparison conditions not established | 0.5833 | [thamilvendhan](https://huggingface.co/datasets/thamilvendhan/signalbench) Reported by a third party | Evaluated revision not stated | 2026-07-08 | | thamilvendhan/signalbench | Task srcMetric srcSetup SRC overall; deterministic action-based grader, no LLM judge; seed 0, n=75Comparison conditions not established | 0.5667 | [signalbench raw per-item responses](https://huggingface.co/datasets/thamilvendhan/signalbench) Reported by a third party | Evaluated revision not stated | 2026-07-08 | | thamilvendhan/signalbench | Task timeMetric timeSetup family=time; n=12Comparison conditions not established | 0.8333 | [thamilvendhan](https://huggingface.co/datasets/thamilvendhan/signalbench) Reported by a third party | Evaluated revision not stated | 2026-07-08 | ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 688.6 GB | | 16-bit | 1369.0 GB | | 8-bit | 684.5 GB | | 4-bit | 342.2 GB | Weights only, from the published parameter count; the key-value cache and runtime add to this. ## Hosted Prices | Host | Input / output | Unit | Observed | | --- | --- | --- | --- | | DeepInfra | $0.32 / $0.89 | input / output, per million tokens | Oct 24, 2026 | From the [SAVRN Index](https://savrn.com/ai-index/pricing/open-models). ## Questions About DeepSeek-V3 ### How much GPU memory does DeepSeek-V3 need? About 1642.8 GB at 16-bit and 410.7 GB at 4-bit: the weights (684.5B parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run DeepSeek-V3 on? At 16-bit, 7x MI325X from $14.00 an hour; at 4-bit, 2x MI325X from $4.00 an hour, at the lowest on-demand prices the SAVRN Index lists. ### What is DeepSeek-V3's context length? 163,840 tokens, from the maximum position embeddings in its published configuration. ## Similar Models Model · Text generation ### [DeepSeek-R1](https://savrn.com/models/deepseek-r1) [DeepSeek](https://savrn.com/model-publishers/deepseek-ai) We introduce our first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1. DeepSeek-R1-Zero, a model trained via large-scale reinforcement learning (RL) without supervised fine-tuning (SFT) as a preliminary step, demonstrated remarkable performance on reasoning. With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors. However, DeepSeek-R1-Zero encounters challenges such as endless repetition, poor readability, and language mixing. To address these issues and further enhance reasoning performance, we introduce DeepSeek-R1, which incorporates cold-start data before RL. DeepSeek-R1 achieves performance comparable to OpenAI-o1 across… Open weights mit 684.5B parameters 163,840 tokens transformers [View model](https://savrn.com/models/deepseek-r1) Model · Text generation ### [DeepSeek-V3-rescaled](https://savrn.com/models/deepseek-v3-rescaled) [Silares](https://savrn.com/model-publishers/sail-ing-north) We present DeepSeek-V3, a strong Mixture-of-Experts (MoE) language model with 671B total parameters with 37B activated for each token. To achieve efficient inference and cost-effective training, DeepSeek-V3 adopts Multi-head Latent Attention (MLA) and DeepSeekMoE architectures, which were thoroughly validated in DeepSeek-V2. Furthermore, DeepSeek-V3 pioneers an auxiliary-loss-free strategy for load balancing and sets a multi-token prediction training objective for stronger performance. We pre-train DeepSeek-V3 on 14.8 trillion diverse and high-quality tokens, followed by Supervised Fine-Tuning and Reinforcement Learning stages to fully harness its capabilities. Comprehensive evaluations… Open weights 684.5B parameters 163,840 tokens transformers [View model](https://savrn.com/models/deepseek-v3-rescaled) Model · Text generation ### [DeepSeek-V3-0324](https://savrn.com/models/deepseek-v3-0324) [DeepSeek](https://savrn.com/model-publishers/deepseek-ai) DeepSeek-V3-0324 demonstrates notable improvements over its predecessor, DeepSeek-V3, in several key aspects. - More aesthetically pleasing web pages and game front-ends - Enhanced report analysis requests with more detailed outputs - Increased accuracy in Function Calling, fixing issues from previous V3 versions In the official DeepSeek web/app, we use the same system prompt with a specific date. For example, In our web and application environments, the temperature parameter $T{model}$ is set to 0.3. Because many users use the default temperature 1.0 in API call, we have implemented an API temperature $T{api}$ mapping mechanism that adjusts the input API temperature value of 1.0 to the… Open weights mit 684.5B parameters 163,840 tokens transformers [View model](https://savrn.com/models/deepseek-v3-0324) Model · Text generation ### [DeepSeek-V3.2](https://savrn.com/models/deepseek-v3-2) [DeepSeek](https://savrn.com/model-publishers/deepseek-ai) We introduce DeepSeek-V3.2, a model that harmonizes high computational efficiency with superior reasoning and agent performance. Our approach is built upon three key technical breakthroughs: 1. DeepSeek Sparse Attention (DSA): We introduce DSA, an efficient attention mechanism that substantially reduces computational complexity while preserving model performance, specifically optimized for long-context scenarios. 2. Scalable Reinforcement Learning Framework: By implementing a robust RL protocol and scaling post-training compute, DeepSeek-V3.2 performs comparably to GPT-5. Notably, our high-compute variant, DeepSeek-V3.2-Speciale, surpasses GPT-5 and exhibits reasoning proficiency on par… Open weights mit 685.4B parameters 163,840 tokens transformers [View model](https://savrn.com/models/deepseek-v3-2) Model · Text generation ### [GLM-5.3](https://savrn.com/models/glm-5-3) [Z.ai](https://savrn.com/model-publishers/zai-org) GLM-5.3 uses the same base model as GLM-5.2 — every gain comes from post-training. Compared with GLM-5.2, it is much better at complex coding and long-horizon tasks: GLM-5.3 supports deployment with the following frameworks. Feel free to try them out: - SGLang — see cookbook - vLLM — see recipes - TokenSpeed — see here - Transformers — see transformers docs - KTransformers — see tutorial - Unsloth — see guide - For deployment on the Ascend NPU platform, inference frameworks such as vLLM-Ascend, xLLM and SGLang are supported — see here. - GLM-5.3 supports controlling the thinking budget through the reasoningeffort parameter, which accepts three levels: low, high, and max. It defaults to max… Open weights other 753.3B parameters 1,048,576 tokens transformers [View model](https://savrn.com/models/glm-5-3) Model · Text generation ### [GLM-5.2-FP8](https://savrn.com/models/glm-5-2-fp8) [Z.ai](https://savrn.com/model-publishers/zai-org) Join our WeChat or Discord community. Check out the GLM-5.2 blog and GLM-5 Technical report. Use GLM-5.2 API services on Z.ai API Platform. Try GLM-5.2 here. [ Paper ] [ GitHub ] We're introducing GLM-5.2, our latest flagship model for long-horizon tasks. It marks a substantial leap in long-horizon task capability over its predecessor GLM-5.1 and, for the first time, delivers that capability on a solid 1M-token context. GLM-5.2's new capabilities include: GLM-5.2 supports deployment with the following frameworks. Feel free to try them out: - SGLang (v0.5.13.post1+) — see cookbook - vLLM (v0.23.0+) — see recipes - Transformers (v0.5.12+) — see transformers docs - KTransformers (v0.5.12+)… Open weights mit 753.3B parameters 1,048,576 tokens transformers [View model](https://savrn.com/models/glm-5-2-fp8) ## DeepSeek [All models and datasets](https://savrn.com/model-publishers/deepseek-ai) ## Versions - [e815299b0bcb](https://savrn.com/models/deepseek-v3/versions/e815299b0bcb) · current 2026-10-24 ## Explore More - [All text generation models](https://savrn.com/models/tasks/text-generation) - [Model comparisons](https://savrn.com/models/comparisons) - [The model directory](https://savrn.com/models) - [Open model prices by host](https://savrn.com/ai-index/pricing/open-models) ## Source - Repository metadata, read 2026-10-19. - [DeepSeek website](https://www.deepseek.com/) - [Hugging Face record](https://huggingface.co/deepseek-ai/DeepSeek-V3) - [How the hub is built](https://savrn.com/model-hub/methodology)