Up to date
deepseek-ai MIT out 19 months ago
685B parameters (8 of 256 experts per token), text in, text out, 160K tokens of context. 4 providers serve it from $0.24 per million input tokens.
Verified 2026-10-25 Rank #57 in LLMs · score 105.7 method v1 Sources 4 Downloads 30d 659.6k · #143 of 619 tracked
Picked by rule from 29 tracked pieces: the lab's own post, the most shared, a field report, the latest.
DeepSeek-V3 is a strong Mixture-of-Experts language model with 671B total parameters and 37B activated for each token.
A series of performance analyses models DeepSeek-V3 training efficiency from theoretical roofline limits to realistic PyTorch profiles.
The 4 picks above, plus these 21. 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.
3 providers with a live price, from $0.24 to $0.29 per million input tokens; the median is $0.25. 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.
Served by 4 API providers; 26 quantized builds for local use; a big base for builders (101 fine-tunes, adapters and merges).
Top Spaces running it: enzostvs/deepsite ♥ 16,604 · openfree/Deepseek-v3-0324-Research ♥ 239 · hadadrjt/ai ♥ 86 · jbilcke-hf/ai-app-factory ♥ 68
Weights only, an estimate: leave room for the KV cache. Method
About 411 GB at INT4, 821 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.
55.1. Each tick is a model on this site with a measured Aider Polyglot row; author-reported numbers stay in Reference. Percent (pass_rate_2). edit_format=diff
Source · leaderboard · 2025-03-24
Weights, every dated article, and repo commits on one rail.
Teal is the weights (2025-03-24), amber triangles are 9 dated articles (each links to it), grey ticks are repo commits.
127 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 Aider Polyglot · Write code with agents | 55.1 percent (pass_rate_2) | edit_format=diff | leaderboard | 2025-03-24 |
| measured SWE-bench Verified · Write code with agents | 42 percent resolved | agent: SWE-Exp | leaderboard | 2025-08-06 |
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 | DeepseekV3ForCausalLM |
| Model type | deepseek_v3 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 3,918,786,560 · F32: 41,555,600 · F8_E4M3: 680,571,043,840 |
| Size band | XXL |
| MoE | yes |
| Experts | 256 |
| Experts per token | 8 |
| Context length | 163,840 (config.json max_position_embeddings) |
| Papers | arXiv:2412.19437 |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · DeepseekV3ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · DeepseekV3ForCausalLM · 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-V3-0324: 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) | 821 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.
quantized: 26finetune: 72merge: 5adapter: 24
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| MaziyarPanahi/DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | mit | MaziyarPanahi | 23 | 2.7M |
| nvidia/DeepSeek-V3-0324-NVFP4 | finetune | NVFP4 | 397B | XXL | mit | nvidia | 22 | 656.3k |
| unsloth/DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | mit | unsloth | 200 | 602.6k |
| unsloth/DeepSeek-V3-0324-BF16 | finetune | -BF16 | 684B | XXL | mit | unsloth | 4 | 200.1k |
| unsloth/DeepSeek-V3-0324-GGUF-UD | quantized | GGUF | — | — | mit | unsloth | 24 | 62.9k |
| bartowski/deepseek-ai_DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | mit | bartowski | 27 | 56.9k |
| QuixiAI/DeepSeek-V3-0324-AWQ | quantized | AWQ | 671B | XXL | mit | QuixiAI | 23 | 37.6k |
| tngtech/DeepSeek-TNG-R1T2-Chimera | merge | — | 684B | XXL | mit | tngtech | 287 | 21k |
| tngtech/DeepSeek-R1T-Chimera | merge | — | 684B | XXL | mit | tngtech | 272 | 14.7k |
| lmstudio-community/DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | mit | lmstudio-community | 12 | 19.2k |
| DevQuasar-12/deepseek-ai.DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | — | DevQuasar-12 | 2 | 11.9k |
| ubergarm/DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | mit | ubergarm | 36 | 6.9k |
| ModelCloud/DeepSeek-V3-0324-BF16 | finetune | -BF16 | 684B | XXL | mit | ModelCloud | 4 | 6.3k |
| unsloth/DeepSeek-V3-0324 | quantized | — | 684B | XXL | mit | unsloth | 8 | 5.8k |
| moxin-org/DeepSeek-V3-0324-Moxin-GGUF | quantized | GGUF | — | — | mit | moxin-org | 8 | 3.9k |
| huihui-ai/DeepSeek-V3-0324-bf16 | finetune | -bf16 | — | — | mit | huihui-ai | 4 | 3.6k |
| mmnga/DeepSeek-V3-0324-experts-pertok-4-gguf | quantized | gguf | — | — | mit | mmnga | 0 | 2.1k |
| bobchenyx/DeepSeek-V3-0324-MLA-GGUF | quantized | GGUF | — | — | mit | bobchenyx | 2 | 2k |
| mlx-community/DeepSeek-V3-0324-5bit | quantized | — | 671B | XXL | mit | mlx-community | 1 | 1.2k |
| tflsxyy/DeepSeek-V3-0324-MoE-Pruner-E160-IQ1_S | quantized | GGUF | — | — | mit | tflsxyy | 4 | 731 |
| gghfez/DeepSeek-V3-0324-256x21B-BF16 | finetune | -BF16 | — | — | — | gghfez | 1 | 784 |
| Kebob/DeepSeek-V3-0324-IK_GGUF | quantized | GGUF | — | — | mit | Kebob | 1 | 680 |
| nebius/MTP-DeepSeek-V3-0324 | quantized | — | 12.5B | L | cc-by-4.0 | nebius | 3 | 501 |
| AmpereComputing/deepseek-v3-0324-gguf | quantized | gguf | — | — | — | AmpereComputing | 0 | 624 |
| RefalMachine/DeepSeek-V3-0324-Channel-INT8 | finetune | INT8 | 685B | XXL | mit | RefalMachine | 2 | 503 |
| BitPhinix/DeepSeek-V3-0324-FP4 | finetune | — | 397B | XXL | mit | BitPhinix | 0 | 588 |
| gghfez/DeepSeek-V3-0324-IQ2_KS | quantized | GGUF | — | — | mit | gghfez | 0 | 572 |
| TVS759/DeepSeek-TNG-R1T2-Chimera | merge | — | 684B | XXL | mit | TVS759 | 0 | 439 |
| gghfez/DeepSeek-V3-0324-IQ3_KS | quantized | GGUF | — | — | mit | gghfez | 2 | 298 |
| jasonyux/DeepSeek-V3-0324-AWQ | quantized | AWQ | 671B | XXL | mit | jasonyux | 0 | 367 |
| lmganon123/DeepSeek-V3-0324_IK_GGUF_Q2 | quantized | GGUF | — | — | mit | lmganon123 | 1 | 216 |
| gghfez/DeepSeek-V3-0324-Opus3-Styletune-GGUF | quantized | GGUF | — | — | cc-by-nc-4.0 | gghfez | 4 | 0 |
| CPU-Hybrid-MoE/DeepSeek-V3-0324-GPU-FP8-GPTQ4 | quantized | -FP8 | 678B | XXL | — | CPU-Hybrid-MoE | 0 | 193 |
| sokann/DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | mit | sokann | 0 | 151 |
| HaveAI/Flare | finetune | — | — | — | mit | HaveAI | 0 | 142 |
| Doctor-Shotgun/DeepSeek-V3-0324-GGUF | quantized | GGUF | — | — | — | Doctor-Shotgun | 0 | 128 |
| CPU-Hybrid-MoE/DeepSeek-V3-0324-CPU-NUMA4-AMXINT8 | finetune | — | — | — | — | CPU-Hybrid-MoE | 0 | 115 |
| CPU-Hybrid-MoE/DeepSeek-V3-0324-CPU-NUMA4-AMXINT4 | finetune | — | — | — | — | CPU-Hybrid-MoE | 0 | 112 |
| Tima5/TimaAI | quantized | — | — | — | apache-2.0 | Tima5 | 0 | 63 |
| fabulouschase/bold | finetune | — | — | — | bigscience-bloom-rail-1.0 | fabulouschase | 1 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2025-03-27 | Small fix | e9b33ad |
| 2025-03-25 | Release DeepSeek-V3-0324 | 500c51d |
| 2025-03-24 | Add files using upload-large-folder tool | f6be68c |
| 2025-03-24 | Add files using upload-large-folder tool | 904b088 |
| 2025-03-24 | Add files using upload-large-folder tool | b984720 |
| 2025-03-24 | Add files using upload-large-folder tool | cb1ae65 |
| 2025-03-24 | Add files using upload-large-folder tool | aa08c59 |
| 2025-03-24 | Add files using upload-large-folder tool | 793d5be |
| 2025-03-24 | Add files using upload-large-folder tool | 4a70c63 |
| 2025-03-24 | Add files using upload-large-folder tool | b9763d9 |
| 2025-03-24 | Add files using upload-large-folder tool | 6364795 |
| 2025-03-24 | initial commit | 586929f |
Hosted endpoints observed for DeepSeek-V3-0324, 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: $0.24/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| DeepInfra hf | deepseek-ai/DeepSeek-V3-0324 | 160K | — | $0.24 | $0.90 | 84.8 | 360 ms | yes | live | 2026-10-25 |
| SiliconFlow openrouter | deepseek/deepseek-chat-v3-0324 | 160K | fp8 | $0.25 | $1.00 | — | — | yes | live | 2026-10-24 |
| GMICloud openrouter | deepseek/deepseek-chat-v3-0324 | 125K | fp8 | $0.29 | $1.14 | — | — | — | live | 2026-10-24 |
| Featherless hf | deepseek-ai/DeepSeek-V3-0324 | — | — | — | — | — | — | — | 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 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.
