Up to date
deepseek-ai MIT out 2 months ago
1.65T parameters (6 of 384 experts per token), text in, text out, 1M tokens of context. 22 providers serve it from $0.12 per million input tokens.
Verified 2026-10-25 Rank #69 in LLMs · score 104.4 method v1 Sources 5 Downloads 30d 151k · #271 of 619 tracked
Picked by rule from 39 tracked pieces: the lab's own post, the most shared, a field report, the latest.
DeepSeek V4 Pro 0813 is a large-scale mixture-of-experts model released on Aug 12, 2026 with a context window of 1.0M.
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.
22 providers with a live price, from $0.12 to $1.65 per million input tokens; the median is $1.07. 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 (22 API providers); 12 quantized builds for local use.
Top Spaces running it: Bit-Trading-Company/ASE-GLM ♥ 2 · mikedorian369/deepseek-v4-pro-0813-showcase ♥ 1 · adamm-hf/P-NP ♥ 1 · rahulthespy/deepseek-v4-pro-0813-demo ♥ 1
Weights only, an estimate: leave room for the KV cache. Method
About 990 GB at INT4, 1.98 TB at FP8, 3.96 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
Claim-matched tests, graded by code. Receipts, not a score.
{
"providers": [
{
"provider": "parasail",
"endpoint_model": "parasail-deepseek-v4-pro-0813",
"source_url": "https://api.parasail.io/v1/chat/completions",
"ttft_ms_median": 1446,
"ttft_any_ms_median": 753,
"tokens_per_s_median": 63.3,
"runs": [
{
"ttft_s": 1.446,
"ttft_any_s": 0.753,
"tokens_per_s": 63.3,
"latency_s": 3.45,
"tokens_out": 218,
"reasoning_chars": 244,
"finish": "stop"
},
{
"ttft_s": 1.384,
"ttft_any_s": 0.759,
"tokens_per_s": 59.6,
"latency_s": 3.58,
"tokens_out": 213,
"reasoning_chars": 244,
"finish": "stop"
},
{
"ttft_s": 1.673,
"ttft_any_s": 0.716,
"tokens_per_s": 66.2,
"latency_s": 3.48,
"tokens_out": 230,
"reasoning_chars": 404,
"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": "The weather in Lisbon on Wednesday 2026-10-14 is clear (no rain expected), so I checked your calendar and found a free evening slot from 19:30–21:30. I've 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 (2026-08-13), amber triangles are 10 dated articles (each links to it), grey ticks are repo commits.
13 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 Agents' Last Exam | 25.7 | — | model author | 2026-10-07 |
| author-reported AutomationBench (Public) | 31.8 | — | model author | 2026-10-07 |
| author-reported Cybergym | 83.3 | — | model author | 2026-10-07 |
| author-reported DSBench-FullStack † | 71.1 | — | model author | 2026-10-07 |
| author-reported DSBench-Hard † | 67.2 | — | model author | 2026-10-07 |
| author-reported DeepSWE · Write code with agents | 62.7 | — | model author | 2026-10-07 |
| author-reported HLE (wo / w tools) | 42.7 / 60.0 other | — | model author | 2026-10-07 |
| author-reported NL2Repo · Write code with agents | 61.5 | — | model author | 2026-10-07 |
| author-reported Terminal-Bench · Write code with agents | 87.9 | — | model author | 2026-10-07 |
| author-reported Toolathlon · Call tools and APIs | 74.1 | — | 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 | DeepseekV4ForCausalLM |
| Model type | deepseek_v4 |
| Dtype | bfloat16 |
| Params by dtype | BF16: 2,954,820,352 · F32: 90,530,058 · F8_E4M3: 23,952,621,568 · I8: 1,623,497,637,888 · I64: 2,327,040 |
| Size band | XXL |
| MoE | yes |
| Experts | 384 |
| Experts per token | 6 |
| Context length | 1,048,576 (config.json max_position_embeddings) |
| Papers | arXiv:2606.19348 |
| License | MIT (mit) |
| Hub snapshot | 2026-10-25 |
vLLM: supported · DeepseekV4ForCausalLM · 7050ef881ff0835f6890a47db1859a24c65481d6SGLang: supported · DeepseekV4ForCausalLM · 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-V4-Pro-0813: params × bytes/param × 1.2 overhead. KV cache and activations are extra. How this is computed.
| BF16 (2 bytes/param) | 3.96 TB · Cluster (≥640 GB) |
|---|---|
| FP8 (1 byte/param) | 1.98 TB · Cluster (≥640 GB) |
| INT4 (0.5 bytes/param) | 990 GB · Cluster (≥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: 12finetune: 1
| Variant | Relation | Quant | Params | Band | License | Author | Likes | Downloads |
|---|---|---|---|---|---|---|---|---|
| unsloth/DeepSeek-V4-Pro-0813-GGUF | quantized | GGUF | — | — | mit | unsloth | 67 | 296.6k |
| 6block/DeepSeek-V4-Pro-0813-GGUF | quantized | GGUF | — | — | mit | 6block | 2 | 18.3k |
| nvidia/DeepSeek-V4-Pro-0813-NVFP4 | quantized | NVFP4 | 1.65T | XXL | mit | nvidia | 20 | 4.8k |
| qtum/DeepSeek-V4-Pro-0813-GGUF | quantized | GGUF | — | — | mit | qtum | 1 | 3.4k |
| nvidia/DeepSeek-V4-Pro-0813-nvfp4-DSpark | quantized | nvfp4 | 1.65T | XXL | mit | nvidia | 9 | 2.4k |
| unsloth/DeepSeek-V4-Pro-0813 | quantized | — | 1.65T | XXL | mit | unsloth | 5 | 1.7k |
| ryant0000/DeepSeek-V4-Pro-0813-gguf | quantized | gguf | — | — | mit | ryant0000 | 0 | 1.6k |
| pokey2025/dspark-DeepSeek-V4-Pro-0813.gguf | quantized | gguf | — | — | mit | pokey2025 | 0 | 742 |
| DevQuasar/deepseek-ai.DeepSeek-V4-Pro-0813-GGUF | quantized | GGUF | — | — | — | DevQuasar | 2 | 473 |
| wrldsuksgo2mars/DeepSeek-V4-Pro-0813-EXL3-K2-calibrated-v1 | quantized | EXL3 | 231B | XXL | mit | wrldsuksgo2mars | 0 | 550 |
| bullerwins/DeepSeek-V4-Pro-0813-exl3-2.04bpw | quantized | exl3 | 217B | XXL | mit | bullerwins | 2 | 302 |
| AxionML/DeepSeek-V4-Pro-0813-NVFP4 | quantized | NVFP4 | 1.65T | XXL | mit | AxionML | 0 | 380 |
| guruswami-ai/deepseek-v4-pro-0813-mlx-tp4-recipe | finetune | mlx | — | — | mit | guruswami-ai | 2 | 0 |
| Date | Commit | SHA |
|---|---|---|
| 2026-08-13 | Update vLLM recipe link (#8) | 72e1d32 |
| 2026-08-13 | Add files using upload-large-folder tool | a2f2780 |
| 2026-08-13 | Add files using upload-large-folder tool | 812a554 |
| 2026-08-13 | Update config.json | 09d796d |
| 2026-08-13 | Release DeepSeek-V4-Pro-0813 | 3c6b304 |
| 2026-08-13 | Add files using upload-large-folder tool | adb0001 |
| 2026-08-13 | Add files using upload-large-folder tool | b689b73 |
| 2026-08-13 | initial commit | 641004c |
Hosted endpoints observed for DeepSeek-V4-Pro-0813, 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: $0.12/M.
| Provider | Model ref | Context | Quant | $/M in | $/M out | tok/s | TTFT | Tools | Status | Observed |
|---|---|---|---|---|---|---|---|---|---|---|
| Relace openrouter | deepseek/deepseek-v4-pro-0813 | 1M | fp4 | $0.12 | $8.00 | — | — | yes | live | 2026-10-24 |
| Baidu openrouter | deepseek/deepseek-v4-pro-0813 | 1M | fp8 | $0.13 | $0.40 | — | — | yes | live | 2026-10-24 |
| Wafer openrouter | deepseek/deepseek-v4-pro-0813 | 1M | mxfp4 | $0.25 | $5.00 | — | — | yes | live | 2026-10-24 |
| Ionstream openrouter | deepseek/deepseek-v4-pro-0813 | 1M | unknown | $0.37 | $2.93 | — | — | yes | live | 2026-10-24 |
| Sail Research openrouter | deepseek/deepseek-v4-pro-0813 | 1M | fp4 | $0.45 | $0.80 | — | — | yes | live | 2026-10-24 |
| Alibaba openrouter | deepseek/deepseek-v4-pro-0813 | 977K | unknown | $0.58 | $1.74 | — | — | yes | live | 2026-10-24 |
| StreamLake openrouter | deepseek/deepseek-v4-pro-0813 | 1000K | unknown | $0.66 | $1.98 | — | — | yes | live | 2026-10-24 |
| DeepSeek openrouter | deepseek/deepseek-v4-pro-0813 | 1M | unknown | $0.66 | $1.98 | — | — | yes | live | 2026-10-24 |
| Phala openrouter | deepseek/deepseek-v4-pro-0813 | 1M | unknown | $0.96 | $2.88 | — | — | yes | live | 2026-10-24 |
| Novita hf, openrouter | deepseek/deepseek-v4-pro-0813 | 1M | fp8 | $0.99 | $2.97 | 48.9 | 1284 ms | yes | live | 2026-10-25 |
| GMICloud openrouter | deepseek/deepseek-v4-pro-0813 | 1024K | fp8 | $1.06 | $3.17 | — | — | yes | live | 2026-10-24 |
| DekaLLM openrouter | deepseek/deepseek-v4-pro-0813 | 1M | unknown | $1.08 | $3.80 | — | — | yes | live | 2026-10-24 |
| DeepInfra hf, deepinfra, openrouter | deepseek-ai/DeepSeek-V4-Pro-0813 | 1M | fp8 | $1.30 | $2.60 | 83.4 | 471 ms | yes | live | 2026-10-25 |
| CoreWeave openrouter | deepseek/deepseek-v4-pro-0813 | 1M | fp8 | $1.31 | $3.96 | — | — | yes | live | 2026-10-24 |
| Together AI hf, openrouter | deepseek-ai/DeepSeek-V4-Pro-0813 | 1M | unknown | $1.32 | $3.96 | 119.2 | 660 ms | yes | live | 2026-10-25 |
| Baseten hf | deepseek-ai/DeepSeek-V4-Pro-0813 | 1M | — | $1.32 | $3.96 | 103.2 | 417 ms | yes | live | 2026-10-25 |
| Parasail parasail, openrouter | parasail-deepseek-v4-pro-0813 | 1M | fp8 | $1.32 | $3.96 | — | — | yes | serverless | 2026-10-24 |
| AtlasCloud openrouter | deepseek/deepseek-v4-pro-0813 | 1M | fp8 | $1.32 | $3.96 | — | — | yes | live | 2026-10-24 |
| DigitalOcean openrouter | deepseek/deepseek-v4-pro-0813 | 1M | unknown | $1.32 | $3.96 | — | — | yes | live | 2026-10-24 |
| SiliconFlow openrouter | deepseek/deepseek-v4-pro-0813 | 1M | fp8 | $1.32 | $3.96 | — | — | yes | live | 2026-10-24 |
| Cloudflare openrouter | deepseek/deepseek-v4-pro-0813 | 1M | unknown | $1.32 | $3.96 | — | — | yes | live | 2026-10-24 |
| Venice openrouter | deepseek/deepseek-v4-pro-0813 | 977K | unknown | $1.65 | $4.95 | — | — | yes | 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.
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 21:33 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.