# DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000 by Diffbot
Source: https://savrn.com/models/deepseek-v4-1-flash-exl3-3bpw-2x-rtx-pro-6000
Markdown alternate of the page above; the site index is https://savrn.com/llms.txt

---

## Runs On

What it takes to serve DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000 (219B 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 | 438.0 GB | 525.6 GB | 2x [MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) (288 GB) Vultr | $5.18 | [3x MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) $5.55 · [3x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $6.00 |
| 8-bit | 219.0 GB | 262.8 GB | 1x [MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) (288 GB) Vultr | $2.59 | [2x MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) $3.70 · [2x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $4.00 |
| 4-bit | 109.5 GB | 131.4 GB | 1x [MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) (192 GB) Vultr | $1.85 | [1x MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) $2.00 · [1x MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) $2.59 |

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 7, 2026.

[DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000 on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/deepseek-v4-1-flash-exl3-3bpw-2x-rtx-pro-6000/gpus)

## Model Card

By Diffbot, published under mit, revision 797f3265e9ed.

### DeepSeek-V4.1-Flash EXL3 3.0 bpw for 2× RTX PRO 6000 Blackwell

[DeepSeek-V4.1-Flash](https://savrn.com/models/deepseek-v4-1-flash), ready to serve on two RTX PRO 6000 Blackwell cards (sm_120, 96 GB each) with vLLM at tensor-parallel 2. The routed experts are [coolbho3k's calibrated EXL3 3.0 bpw quantization](https://huggingface.co/coolbho3k/DeepSeek-V4.1-Flash-EXL3-3bpw). Our additions are:

- int4 Engram tables;
- a 3-bit DSpark drafter;
- a serving stack that keeps 30% of the routed experts (the ones agentic coding uses least) in pinned host RAM and runs them next to the VRAM experts;
- a decode-once prefill kernel for the 3-bit experts.

The recipe/ folder has the image build, the sm_120 patches, the MoE kernel, and the serving, benchmark, KL and quantization scripts.

[Read the full model card (1,674 words)](https://savrn.com/models/deepseek-v4-1-flash-exl3-3bpw-2x-rtx-pro-6000/card)

## Configuration

Architecture

DeepseekV41ForCausalLM

Context length (tokens)

1,048,576

Layers

40

Hidden size

5,120

Attention heads

64

Key/value heads

1

Head dimension

512

Vocabulary size

129,280

Routed experts

384

Experts active per token

6

Sliding window (tokens)

128

RoPE base

10,000

Model type

deepseek_v41

Quantization

exl3

## Identity and Version

Repository

diffbot/DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000

Publisher

Diffbot

Task

Text generation

Modality

Text

Library

Not stated by the source

Parameters

219B parameters

Languages

Not stated by the source

Revision

797f3265e9edb5e9d5a0bcae89e011e1e2517215

First published

2026-09-11

Last updated

2026-10-02

## Files and Weights

181 files, 332.1 GB in total. The weights are 56 files totalling 332.1 GB in safetensors.

Weights56 files · 332.1 GB

Configuration45 files · 15.1 MB

Tokenizer2 files · 6.4 MB

Documentation17 files · 109.8 KB

Other59 files · 3.0 MB

Repository2 files · 2.0 KB

Every file

| File | Type | Size | SHA-256 |
| --- | --- | --- | --- |
| model-00001-of-00056.safetensors | Weights | 4.3 GB | 0823259177aa |
| model-00002-of-00056.safetensors | Weights | 4.3 GB | 2267dda5f372 |
| model-00003-of-00056.safetensors | Weights | 4.3 GB | bb473ba21d10 |
| model-00004-of-00056.safetensors | Weights | 4.3 GB | 6867a3737ec3 |
| model-00005-of-00056.safetensors | Weights | 4.3 GB | 4b7c5764931a |
| model-00006-of-00056.safetensors | Weights | 4.3 GB | ba042ab360ab |
| model-00007-of-00056.safetensors | Weights | 4.3 GB | d281e20701bf |
| model-00008-of-00056.safetensors | Weights | 4.3 GB | b79f86d323fd |
| model-00009-of-00056.safetensors | Weights | 4.3 GB | 358e3077afa3 |
| model-00010-of-00056.safetensors | Weights | 4.3 GB | ac00d1064b69 |
| model-00011-of-00056.safetensors | Weights | 4.3 GB | 523f41363278 |
| model-00012-of-00056.safetensors | Weights | 4.3 GB | 57c837cd7a3b |
| model-00013-of-00056.safetensors | Weights | 4.3 GB | 4f2e00a289fb |
| model-00014-of-00056.safetensors | Weights | 4.3 GB | f17863dd245f |
| model-00015-of-00056.safetensors | Weights | 4.3 GB | fdf301fa38bd |
| model-00016-of-00056.safetensors | Weights | 4.3 GB | 3190a7619242 |
| model-00017-of-00056.safetensors | Weights | 4.3 GB | 4b65ea04fc9d |
| model-00018-of-00056.safetensors | Weights | 4.3 GB | 9732aa130aa4 |
| model-00019-of-00056.safetensors | Weights | 4.3 GB | 1e524d3c217a |
| model-00020-of-00056.safetensors | Weights | 4.3 GB | b774f63e093d |
| model-00021-of-00056.safetensors | Weights | 4.3 GB | b6e24b14f6c1 |
| model-00022-of-00056.safetensors | Weights | 4.3 GB | c79c6965a84b |
| model-00023-of-00056.safetensors | Weights | 4.3 GB | 8f1eec728024 |
| model-00024-of-00056.safetensors | Weights | 4.3 GB | 898c41fef3a0 |
| model-00025-of-00056.safetensors | Weights | 4.3 GB | 08b825a8b4bb |
| model-00026-of-00056.safetensors | Weights | 4.3 GB | 5c3ba30481b9 |
| model-00027-of-00056.safetensors | Weights | 4.3 GB | a43a8e9e0045 |
| model-00028-of-00056.safetensors | Weights | 4.3 GB | 83cb70fa8ea8 |
| model-00029-of-00056.safetensors | Weights | 4.3 GB | ec18bf3170b4 |
| model-00030-of-00056.safetensors | Weights | 4.3 GB | f806982b553c |
| model-00031-of-00056.safetensors | Weights | 4.3 GB | 21d92ecc236f |
| model-00032-of-00056.safetensors | Weights | 4.3 GB | 9bb721169be3 |
| model-00033-of-00056.safetensors | Weights | 4.3 GB | 33826e4c2d61 |
| model-00034-of-00056.safetensors | Weights | 4.3 GB | c088432b9049 |
| model-00035-of-00056.safetensors | Weights | 4.3 GB | 04e40c163580 |
| model-00036-of-00056.safetensors | Weights | 4.3 GB | cf35c9f97093 |
| model-00037-of-00056.safetensors | Weights | 4.3 GB | 9f2557ffbc2d |
| model-00038-of-00056.safetensors | Weights | 4.3 GB | 156c74b5dac2 |
| model-00039-of-00056.safetensors | Weights | 4.3 GB | cbab5210abf7 |
| model-00040-of-00056.safetensors | Weights | 4.3 GB | 027b82e21d4a |
| model-00041-of-00056.safetensors | Weights | 4.3 GB | 1038eb0c0f07 |
| model-00042-of-00056.safetensors | Weights | 4.3 GB | e0ed09100cbf |
| model-00043-of-00056.safetensors | Weights | 4.3 GB | 9638aeeb100e |
| model-00044-of-00056.safetensors | Weights | 4.3 GB | 2bcca72b9ad6 |
| model-00045-of-00056.safetensors | Weights | 4.3 GB | 4672f5121172 |
| model-00046-of-00056.safetensors | Weights | 4.3 GB | 15e5b0f7de86 |
| model-00047-of-00056.safetensors | Weights | 4.3 GB | 0fe8c800cb4d |
| model-00048-of-00056.safetensors | Weights | 4.3 GB | a53dbf59b94f |
| model-00049-of-00056.safetensors | Weights | 4.3 GB | c8d171c694b4 |
| model-00050-of-00056.safetensors | Weights | 4.3 GB | 8bdc9d9f802d |
| model-00051-of-00056.safetensors | Weights | 686.8 MB | 3286b6712dae |
| model-00052-of-00056.safetensors | Weights | 2.0 GB | ef469676eb7e |
| model-00053-of-00056.safetensors | Weights | 1.9 GB | c678e4facf45 |
| model-00054-of-00056.safetensors | Weights | 2.0 GB | 2c1a8e9f4b0c |
| model-00055-of-00056.safetensors | Weights | 55.5 GB | 91432b7043f9 |
| model-00056-of-00056.safetensors | Weights | 55.5 GB | 9d21e20e5ddf |
| config.json | Configuration | 4.5 KB | — |
| encoding/encoding.py | Configuration | 37.3 KB | — |
| encoding/test_encoding.py | Configuration | 19.4 KB | — |
| encoding/tests/test_input_1.json | Configuration | 2.8 KB | — |
| encoding/tests/test_input_2.json | Configuration | 527 B | — |
| encoding/tests/test_input_3.json | Configuration | 2.6 KB | — |
| encoding/tests/test_input_4.json | Configuration | 712 B | — |
| encoding/tests/test_input_5.json | Configuration | 1.1 KB | — |
| inference/config.json | Configuration | 2.0 KB | — |
| inference/convert.py | Configuration | 9.5 KB | — |
| inference/engram.py | Configuration | 8.1 KB | — |
| inference/examples/example_harmony.json | Configuration | 2.2 KB | — |
| inference/generate.py | Configuration | 8.7 KB | — |
| inference/image_processor.py | Configuration | 7.7 KB | — |
| inference/kernel.py | Configuration | 23.8 KB | — |
| inference/model.py | Configuration | 61.5 KB | — |
| inference/vision.py | Configuration | 4.5 KB | — |
| model.safetensors.index.json | Configuration | 14.6 MB | 743e9baf5e37 |
| recipe/bench/bench-fanout.py | Configuration | 3.1 KB | — |
| recipe/bench/bench-sbs.py | Configuration | 2.6 KB | — |
| recipe/bench/greedy-ref.py | Configuration | 1.4 KB | — |
| recipe/bench/hit_bench.py | Configuration | 5.2 KB | — |
| recipe/bench/korean_check.py | Configuration | 4.1 KB | — |
| recipe/bench/len_sweep.py | Configuration | 1.4 KB | — |
| recipe/bench/make_corpus.py | Configuration | 961 B | — |
| recipe/bench/needle-test.py | Configuration | 1.5 KB | — |
| recipe/bench/quality-probe.py | Configuration | 4.3 KB | — |
| recipe/bench/test-toolcall.py | Configuration | 1.9 KB | — |
| recipe/bench/trace_breakdown.py | Configuration | 2.1 KB | — |
| recipe/kernels/build_xmoe.py | Configuration | 904 B | — |
| recipe/kernels/build_xmoe_do.py | Configuration | 767 B | — |
| recipe/kernels/thor_dense.py | Configuration | 10.2 KB | — |
| recipe/kernels/thor_dense2.py | Configuration | 14.3 KB | — |
| recipe/kernels/thor_dense_table_v2.json | Configuration | 1.5 KB | — |
| recipe/kl/capture_server.py | Configuration | 3.6 KB | — |
| recipe/patches/vllm/sitecustomize.py | Configuration | 56.1 KB | — |
| recipe/patches/vllm/test_dsml_nan_args.py | Configuration | 1.1 KB | — |
| recipe/plans/make_plan.py | Configuration | 3.0 KB | — |
| recipe/plans/plan-agent30-dh.json | Configuration | 24.4 KB | — |
| recipe/plans/plan-blend30-dh.json | Configuration | 24.4 KB | — |
| recipe/plans/plan-gen30-dh.json | Configuration | 24.4 KB | — |
| recipe/quantize/assemble_pack.py | Configuration | 2.9 KB | — |
| recipe/quantize/convert_engram_int4.py | Configuration | 3.5 KB | — |
| recipe/quantize/quantize_mtp_exl3.py | Configuration | 22.4 KB | — |
| recipe/results/raw/korean_check.json | Configuration | 80.2 KB | — |
| LICENSE | Documentation | 1.1 KB | — |
| README.md | Documentation | 11.8 KB | — |
| encoding/README.md | Documentation | 12.1 KB | — |
| evaluation/README.md | Documentation | 4.0 KB | — |
| inference/README.md | Documentation | 2.0 KB | — |
| recipe/LICENSE | Documentation | 1.1 KB | — |
| recipe/MODEL_CARD.md | Documentation | 11.8 KB | — |
| recipe/README.md | Documentation | 11.1 KB | — |
| recipe/THIRD_PARTY_NOTICES.md | Documentation | 3.2 KB | — |
| recipe/docker/LICENSE.sfxnz-recipe | Documentation | 1.1 KB | — |
| recipe/kernels/xmoe/LICENSE.exllamav3 | Documentation | 1.1 KB | — |
| recipe/kl/README.md | Documentation | 2.0 KB | — |
| recipe/patches/vllm-exl3/LICENSE | Documentation | 31.7 KB | — |
| recipe/patches/vllm-exl3/README.md | Documentation | 2.3 KB | — |
| recipe/plans/README.md | Documentation | 3.2 KB | — |
| recipe/quantize/README.md | Documentation | 2.8 KB | — |
| recipe/results/RESULTS.md | Documentation | 7.6 KB | — |
| DeepSeek_V41_Tech_Report.pdf | Other | 1.8 MB | ba68e2e40408 |
| assets/dsv41_agentic_performance.png | Other | 190.7 KB | 44deae01cb9c |
| assets/dsv41_kv_cache.png | Other | 270.9 KB | b61bf4651d4b |
| encoding/tests/test_output_1.txt | Other | 2.5 KB | — |
| encoding/tests/test_output_2.txt | Other | 294 B | — |
| encoding/tests/test_output_3.txt | Other | 2.5 KB | — |
| encoding/tests/test_output_4.txt | Other | 574 B | — |
| encoding/tests/test_output_5.txt | Other | 408 B | — |
| evaluation/dsh-minimal.patch | Other | 28.7 KB | — |
| inference/examples/example.txt | Other | 332 B | — |
| inference/examples/images/carrots.jpeg | Other | 212.5 KB | 5df896a4a07e |
| inference/examples/images/corn.jpeg | Other | 56.1 KB | — |
| inference/requirements.txt | Other | 97 B | — |
| inference/run.sh | Other | 1.8 KB | — |
| recipe/bench/ab-run.sh | Other | 2.1 KB | — |
| recipe/bench/fetch-evaldata.sh | Other | 1.1 KB | — |
| recipe/bench/profile-run.sh | Other | 2.5 KB | — |
| recipe/bench/validate.sh | Other | 1.7 KB | — |
| recipe/docker/build.sh | Other | 4.3 KB | — |
| recipe/kernels/build.sh | Other | 794 B | — |
| recipe/kernels/exllamav3-5be8865-exl3_moe.patch | Other | 82.6 KB | — |
| recipe/kernels/xmoe/bindings.cpp | Other | 726 B | — |
| recipe/kernels/xmoe/compat.cuh | Other | 642 B | — |
| recipe/kernels/xmoe/graph.cuh | Other | 3.0 KB | — |
| recipe/kernels/xmoe/ptx.cuh | Other | 9.6 KB | — |
| recipe/kernels/xmoe/quant/codebook.cuh | Other | 5.9 KB | — |
| recipe/kernels/xmoe/quant/comp_units/exl3_moe_inst_k2_cb1.cu | Other | 654 B | — |
| recipe/kernels/xmoe/quant/comp_units/exl3_moe_inst_k3_cb1.cu | Other | 654 B | — |
| recipe/kernels/xmoe/quant/comp_units/exl3_moe_inst_k3_cb2.cu | Other | 576 B | — |
| recipe/kernels/xmoe/quant/comp_units/exl3_moe_instances.cuh | Other | 1.3 KB | — |
| recipe/kernels/xmoe/quant/exl3_devctx.cu | Other | 2.9 KB | — |
| recipe/kernels/xmoe/quant/exl3_devctx.cuh | Other | 1.6 KB | — |
| recipe/kernels/xmoe/quant/exl3_dq.cuh | Other | 10.3 KB | — |
| recipe/kernels/xmoe/quant/exl3_gemm.cuh | Other | 2.2 KB | — |
| recipe/kernels/xmoe/quant/exl3_gemm_inner.cuh | Other | 23.9 KB | — |
| recipe/kernels/xmoe/quant/exl3_gemm_inner_flat.cuh | Other | 14.9 KB | — |
| recipe/kernels/xmoe/quant/exl3_gemm_inner_mt.cuh | Other | 24.0 KB | — |
| recipe/kernels/xmoe/quant/exl3_kernel_map.cuh | Other | 4.8 KB | — |
| recipe/kernels/xmoe/quant/exl3_moe.cu | Other | 15.2 KB | — |
| recipe/kernels/xmoe/quant/exl3_moe.cuh | Other | 1.2 KB | — |
| recipe/kernels/xmoe/quant/exl3_moe_common.cuh | Other | 2.4 KB | — |
| recipe/kernels/xmoe/quant/exl3_moe_flat_kernel.cuh | Other | 8.6 KB | — |
| recipe/kernels/xmoe/quant/exl3_moe_kernel.cuh | Other | 14.4 KB | — |
| recipe/kernels/xmoe/quant/hadamard_inner.cuh | Other | 12.8 KB | — |
| recipe/kernels/xmoe/util.cuh | Other | 4.7 KB | — |
| recipe/kernels/xmoe/util.h | Other | 8.8 KB | — |
| recipe/kernels/xmoe_do/do_gemm.cu | Other | 12.2 KB | — |
| recipe/patches/vllm-exl3/NOTICE.vllm-exl3 | Other | 989 B | — |
| recipe/patches/vllm-exl3/exl3-thor.patch | Other | 48.1 KB | — |
| recipe/patches/vllm/pr58132-decoder-swa-replay.diff | Other | 35.2 KB | — |
| recipe/patches/vllm/pr59585-dsml-finite-args.diff | Other | 1.8 KB | — |
| recipe/patches/vllm/sm120-thor-fixes.diff | Other | 13.5 KB | — |
| recipe/quantize/quantize_mtp_exl3.diff | Other | 6.0 KB | — |
| recipe/results/raw/kl-served-1001b.txt | Other | 293 B | — |
| recipe/results/raw/kl-served-rope-bug.txt | Other | 269 B | — |
| recipe/results/raw/kl-served.txt | Other | 269 B | — |
| recipe/results/raw/validate-recipe-1001b.txt | Other | 2.8 KB | — |
| recipe/results/raw/validate-recipe.txt | Other | 4.1 KB | — |
| recipe/serve/serve.sh | Other | 9.3 KB | — |
| .gitattributes | Repository | 1.9 KB | — |
| recipe/.gitignore | Repository | 143 B | — |
| tokenizer.json | Tokenizer | 6.4 MB | — |
| tokenizer_config.json | Tokenizer | 801 B | — |

## License and Download

License

mit

Access

Open weights, no gate

Download size

332.1 GB

[Download from Diffbot](https://huggingface.co/diffbot/DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000)

Released by Diffbot through its official repository on Hugging Face. [Read the license](https://opensource.org/license/mit).

## Built From

- Derived from [deepseek-ai/DeepSeek-V4.1-Flash](https://savrn.com/models/deepseek-v4-1-flash)
- Quantized from [deepseek-ai/DeepSeek-V4.1-Flash](https://savrn.com/models/deepseek-v4-1-flash)

## Memory Requirements

| Precision | Weights in memory |
| --- | --- |
| As published | 332.1 GB |
| 16-bit | 438.0 GB |
| 8-bit | 219.0 GB |
| 4-bit | 109.5 GB |

Weights only, from the published parameter count; the key-value cache and runtime add to this.

## Questions About DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000

### How much GPU memory does DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000 need?

About 525.6 GB at 16-bit and 131.4 GB at 4-bit: the weights (219B parameters) plus a working margin. A long context needs more.

### What is the cheapest GPU to run DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000 on?

At 16-bit, 2x MI355X from $5.18 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

### Can I use DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000 commercially?

Yes. DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000 is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.

### What is DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000's context length?

1,048,576 tokens, from the maximum position embeddings in its published configuration.

## Similar Models

Model · Text generation

### [Hy3-Razor-226B-A18B-E144of192](https://savrn.com/models/hy3-razor-226b-a18b-e144of192)

[Mingyang Song](https://savrn.com/model-publishers/nickyang)

Hy3 with a quarter of its routed experts removed by RAZOR, a training-free expert pruning method. Every MoE layer keeps 144 of its original 192 routed experts. No gradient updates or recovery training were applied: the retained weights are the base model's own weights. Pruning touches only the routed expert pool. Attention, shared experts, the embedding and the LM head are untouched, so the compute per token drops only by the share of expert FLOPs that the removed experts would have contributed. The other budget is Requires a Transformers build containing the native hyv3 implementation. Weights are bfloat16. RAZOR asks whether the surviving computation can replace an expert's function…

Open weights apache-2.0 226.3B parameters 262,144 tokens transformers

[View model](https://savrn.com/models/hy3-razor-226b-a18b-e144of192)

Model · Text generation

### [MiniMax-M2.7](https://savrn.com/models/minimax-m2-7)

[MiniMax](https://savrn.com/model-publishers/minimaxai)

Join Our WeChat Discord community. MiniMax Agent API CLI MiniMax Website Hugging Face GitHub ModelScope LICENSE MiniMax-M2.7 is our first model deeply participating in its own evolution. M2.7 is capable of building complex agent harnesses and completing highly elaborate productivity tasks, leveraging Agent Teams, complex Skills, and dynamic tool search. For more details, see our blog post. M2.7 initiates a cycle of model self-evolution: during development, we let the model update its own memory, build dozens of complex skills for RL experiments, and improve its own learning process based on experiment results. An internal version of M2.7 autonomously optimized a programming scaffold over…

Open weights other 228.7B parameters 204,800 tokens transformers

[View model](https://savrn.com/models/minimax-m2-7)

Model · Text generation

### [Qwen3-235B-A22B-Thinking-2507](https://savrn.com/models/qwen3-235b-a22b-thinking-2507)

[Zachary Howard](https://savrn.com/model-publishers/servantofares)

Over the past three months, we have continued to scale the thinking capability of Qwen3-235B-A22B, improving both the quality and depth of reasoning. We are pleased to introduce Qwen3-235B-A22B-Thinking-2507, featuring the following key enhancements: - Significantly improved performance on reasoning tasks, including logical reasoning, mathematics, science, coding, and academic benchmarks that typically require human expertise — achieving state-of-the-art results among open-source thinking models. - Markedly better general capabilities, such as instruction following, tool usage, text generation, and alignment with human preferences. - Enhanced 256K long-context understanding capabilities.…

Open weights apache-2.0 235.1B parameters 262,144 tokens transformers

[View model](https://savrn.com/models/qwen3-235b-a22b-thinking-2507)

Model · Text generation

### [DeepSeek-V4-Flash-DSpark](https://savrn.com/models/deepseek-v4-flash-dspark)

[DeepSeek](https://savrn.com/model-publishers/deepseek-ai)

Note: DeepSeek-V4-Flash-DSpark is not a new model. It is the same checkpoint with an additional speculative decoding module attached. A minimal inference example is available in the inference folder. For more details, refer to: https://github.com/deepseek-ai/DeepSpec We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models — DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens. DeepSeek-V4 series incorporate several key upgrades in architecture and optimization: 1. Hybrid Attention Architecture: We design a hybrid…

Open weights mit 165.3B parameters 1,048,576 tokens transformers

[View model](https://savrn.com/models/deepseek-v4-flash-dspark)

Model · Text generation

### [MiMo-V2.6-Flash-RL](https://savrn.com/models/mimo-v2-6-flash-rl)

[Xiaomi MiMo](https://savrn.com/model-publishers/xiaomimimo)

Scaling Reinforcement Learning Toward Self-Improvement MiMo-V2.6-Flash-RL is the efficiency-balanced checkpoint of the MiMo-V2.6 series. The series is built to scale reinforcement learning toward self-improvement — scaling RL compute, environment diversity, and grader compute together, so the model keeps expanding its capability frontier through exploration and feedback. Key features include: - Native Omnimodal + Long Horizon: Text, image, video, and audio in one model; 1M tokens for long repositories, tool traces, and multi-session agent runs. - Multi-Prefix Multi-Teacher On-Policy Distillation (MOPD2): After mixed RL, MOPD2 combines autonomous student rollouts with prefix-conditioned…

Open weights mit 159.4B parameters 1,048,576 tokens transformers

[View model](https://savrn.com/models/mimo-v2-6-flash-rl)

Model · Text generation

### [MiMo-V2.6-Flash-RL](https://savrn.com/models/mimo-v2-6-flash-rl-2)

[Zachary Howard](https://savrn.com/model-publishers/servantofares)

Scaling Reinforcement Learning Toward Self-Improvement MiMo-V2.6-Flash-RL is the efficiency-balanced checkpoint of the MiMo-V2.6 series. The series is built to scale reinforcement learning toward self-improvement — scaling RL compute, environment diversity, and grader compute together, so the model keeps expanding its capability frontier through exploration and feedback. Key features include: - Native Omnimodal + Long Horizon: Text, image, video, and audio in one model; 1M tokens for long repositories, tool traces, and multi-session agent runs. - Multi-Prefix Multi-Teacher On-Policy Distillation (MOPD2): After mixed RL, MOPD2 combines autonomous student rollouts with prefix-conditioned…

Open weights mit 159.4B parameters 1,048,576 tokens transformers

[View model](https://savrn.com/models/mimo-v2-6-flash-rl-2)

## Diffbot

[All models and datasets](https://savrn.com/model-publishers/diffbot)

## Versions

- [797f3265e9ed](https://savrn.com/models/deepseek-v4-1-flash-exl3-3bpw-2x-rtx-pro-6000/versions/797f3265e9ed) · current 2026-10-02

## Explore More

- [All text generation models](https://savrn.com/models/tasks/text-generation)
- [All models under mit](https://savrn.com/models/licenses/mit)
- [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 2026-10-02.
- [Hugging Face record](https://huggingface.co/diffbot/DeepSeek-V4.1-Flash-EXL3-3bpw-2x-RTX-PRO-6000)
- [How the hub is built](https://savrn.com/model-hub/methodology)