# Inkling by Thinking Machines Lab: Open-Weight Model Source: https://savrn.com/models/inkling Markdown alternate of the page above; the site index is https://savrn.com/llms.txt --- ## Runs On What it takes to serve Inkling (952.4B 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 | 1904.8 GB | 2285.7 GB | 8x [MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) (288 GB) Vultr | $20.72 | | | 8-bit | 952.4 GB | 1142.9 GB | 5x [MI325X](https://savrn.com/ai-index/pricing/gpus/mi325x) (256 GB) Vultr | $10.00 | [4x MI355X](https://savrn.com/ai-index/pricing/gpus/mi355x) $10.36 · [6x MI300X](https://savrn.com/ai-index/pricing/gpus/mi300x) $11.10 | | 4-bit | 476.2 GB | 571.4 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 | 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 12, 2026. [Inkling on every accelerator the SAVRN Index prices, at every precision](https://savrn.com/models/inkling/gpus) ## Model Card By Thinking Machines Lab, published under apache-2.0, revision 828496eeae4c. [BF16](https://savrn.com/models/inkling) | [NVFP4](https://huggingface.co/thinkingmachines/Inkling-NVFP4) | [Playground](https://tinker.thinkingmachines.ai/playground) | [Tinker Cookbook](https://github.com/thinking-machines-lab/tinker-cookbook) | [Acceptable Use](https://thinkingmachines.ai/model-acceptable-use-policy) ### 1. General Information Inkling is a general-purpose multimodal model that accepts text, image and audio inputs and generates text outputs. It is intended for use in English and other languages, and across multiple coding languages. The model is designed to be used by developers building AI-powered applications, including agentic and tool-use systems, coding assistants, chatbots, and retrieval-augmented generation systems, and is suitable for general-purpose conversational use, instruction-following, and other natural language and multimodal tasks. It is released with open weights to support research, fine-tuning and integration into third-party products by downstream developers. Languages: English, with general multilingual capabilities across other languages. ### 2. Getting Started Try Inkling on the [Tinker Playground](https://tinker.thinkingmachines.ai/playground) or access via API using the [Tinker Cookbook](https://github.com/thinking-machines-lab/tinker-cookbook). Inkling supports local deployment using the following open-source libraries: - SGLang ([recipe](https://docs.sglang.io/cookbook/autoregressive/ThinkingMachines/Inkling)) - vLLM ([recipe](https://recipes.vllm.ai/thinkingmachines/Inkling)) - TokenSpeed ([recipe](https://lightseek.org/tokenspeed/recipes/models#Inkling)) - Unsloth ([recipe](https://unsloth.ai/docs/models/inkling)) - Huggingface ([recipe](https://hf.co/blog/thinkingmachines-inkling)) [Read the full model card (1,241 words)](https://savrn.com/models/inkling/card) ## Configuration Architecture InklingForConditionalGeneration Layers 66 Hidden size 6,144 Feed-forward size 3,072 Attention heads 64 Key/value heads 8 Head dimension 128 Vocabulary size 201,024 Routed experts 256 Experts active per token 6 Stored precision bfloat16 Model type inkling_mm_model ## Identity and Version Repository thinkingmachines/Inkling Publisher Thinking Machines Lab Task Image and text to text Modality Image and text Library transformers Parameters 952.4B parameters Languages moe Revision 828496eeae4c243ff1a22f7f28ff83694f2f7bc9 First published 2026-07-14 Last updated 2026-07-23 ## Files and Weights 125 files, 1.9 TB in total. The weights are 109 files totalling 1.9 TB in safetensors. Weights109 files · 1.9 TB Configuration10 files · 133.7 KB Tokenizer3 files · 31.5 MB Documentation1 file · 14.0 KB Other1 file · 6.3 KB Repository1 file · 7.6 KB Every file | File | Type | Size | SHA-256 | | --- | --- | --- | --- | | model-00001-of-00108.safetensors | Weights | 20.0 GB | 9c388741a7b7 | | model-00002-of-00108.safetensors | Weights | 10.6 GB | 5e574d46e46e | | model-00003-of-00108.safetensors | Weights | 19.8 GB | 06fdee46007c | | model-00004-of-00108.safetensors | Weights | 19.6 GB | 2fcc532de433 | | model-00005-of-00108.safetensors | Weights | 19.4 GB | 5be72f83b0e7 | | model-00006-of-00108.safetensors | Weights | 19.6 GB | 70c922e6066b | | model-00007-of-00108.safetensors | Weights | 19.5 GB | 68cbd6ff4319 | | model-00008-of-00108.safetensors | Weights | 19.3 GB | d93c3c5e76e3 | | model-00009-of-00108.safetensors | Weights | 9.8 GB | 99499011c6b4 | | model-00010-of-00108.safetensors | Weights | 19.6 GB | 1b1cf627b6cf | | model-00011-of-00108.safetensors | Weights | 20.0 GB | 130bf8cf37a4 | | model-00012-of-00108.safetensors | Weights | 19.3 GB | 263060235dc3 | | model-00013-of-00108.safetensors | Weights | 9.7 GB | c3bc3a53ea5c | | model-00014-of-00108.safetensors | Weights | 19.3 GB | a9473d1ca561 | | model-00015-of-00108.safetensors | Weights | 19.4 GB | 468c392ae2f2 | | model-00016-of-00108.safetensors | Weights | 19.7 GB | c25da56bd3bf | | model-00017-of-00108.safetensors | Weights | 9.8 GB | e9c9d7e1b787 | | model-00018-of-00108.safetensors | Weights | 19.7 GB | 34f31b3d31f6 | | model-00019-of-00108.safetensors | Weights | 9.7 GB | 3db4a1ba8d71 | | model-00020-of-00108.safetensors | Weights | 19.6 GB | ca12bcd167e6 | | model-00021-of-00108.safetensors | Weights | 19.9 GB | e1f90bd55136 | | model-00022-of-00108.safetensors | Weights | 10.2 GB | 967b38c7b9ec | | model-00023-of-00108.safetensors | Weights | 19.8 GB | 1c07b7c00891 | | model-00024-of-00108.safetensors | Weights | 19.5 GB | f8f6bd899184 | | model-00025-of-00108.safetensors | Weights | 19.7 GB | 17fb32b0a565 | | model-00026-of-00108.safetensors | Weights | 20.0 GB | 05592903630c | | model-00027-of-00108.safetensors | Weights | 19.5 GB | e77b18cf34b4 | | model-00028-of-00108.safetensors | Weights | 19.8 GB | 00f9a91b444a | | model-00029-of-00108.safetensors | Weights | 19.6 GB | d8d969ea57bd | | model-00030-of-00108.safetensors | Weights | 19.3 GB | cce7322e73a1 | | model-00031-of-00108.safetensors | Weights | 10.0 GB | 639b6a16bcea | | model-00032-of-00108.safetensors | Weights | 19.9 GB | 38dd3294bb46 | | model-00033-of-00108.safetensors | Weights | 19.5 GB | 24ccc4eabdc4 | | model-00034-of-00108.safetensors | Weights | 19.5 GB | 233192acd04d | | model-00035-of-00108.safetensors | Weights | 19.3 GB | c6440b8a8692 | | model-00036-of-00108.safetensors | Weights | 19.9 GB | 6132edf471e2 | | model-00037-of-00108.safetensors | Weights | 19.9 GB | 2c01176c8419 | | model-00038-of-00108.safetensors | Weights | 19.4 GB | cbcdc896300e | | model-00039-of-00108.safetensors | Weights | 19.4 GB | 6d2456e7b9ba | | model-00040-of-00108.safetensors | Weights | 19.7 GB | 5a50326d2a5e | | model-00041-of-00108.safetensors | Weights | 19.5 GB | 0c2d086d21f9 | | model-00042-of-00108.safetensors | Weights | 19.9 GB | 1bd33ab40986 | | model-00043-of-00108.safetensors | Weights | 19.6 GB | 1e0a1e4c835f | | model-00044-of-00108.safetensors | Weights | 13.0 GB | 46053daa0b51 | | model-00045-of-00108.safetensors | Weights | 19.9 GB | 79448cab9cd4 | | model-00046-of-00108.safetensors | Weights | 11.2 GB | 11125c58e321 | | model-00047-of-00108.safetensors | Weights | 19.3 GB | 424a1d85c9f8 | | model-00048-of-00108.safetensors | Weights | 19.6 GB | ca1d77da444e | | model-00049-of-00108.safetensors | Weights | 20.0 GB | 14c57a2db7c7 | | model-00050-of-00108.safetensors | Weights | 10.1 GB | 3d4922745360 | | model-00051-of-00108.safetensors | Weights | 19.4 GB | 0b431ec843d5 | | model-00052-of-00108.safetensors | Weights | 10.7 GB | 76466f8dc307 | | model-00053-of-00108.safetensors | Weights | 19.7 GB | e78eb2cc948f | | model-00054-of-00108.safetensors | Weights | 19.4 GB | 6484f02b59a7 | | model-00055-of-00108.safetensors | Weights | 717.2 MB | 455afae3aba5 | | model-00056-of-00108.safetensors | Weights | 19.4 GB | c16e70640f42 | | model-00057-of-00108.safetensors | Weights | 19.6 GB | 9936ebd4c4ac | | model-00058-of-00108.safetensors | Weights | 19.6 GB | b56666517759 | | model-00059-of-00108.safetensors | Weights | 20.0 GB | ba23d7a90e70 | | model-00060-of-00108.safetensors | Weights | 19.5 GB | c878905e47fe | | model-00061-of-00108.safetensors | Weights | 19.6 GB | d604321f1be9 | | model-00062-of-00108.safetensors | Weights | 19.4 GB | e977b6b7eb30 | | model-00063-of-00108.safetensors | Weights | 19.3 GB | bb6f1c65b265 | | model-00064-of-00108.safetensors | Weights | 19.4 GB | 2a04c35fdd82 | | model-00065-of-00108.safetensors | Weights | 19.4 GB | 9cee5706b291 | | model-00066-of-00108.safetensors | Weights | 19.3 GB | 84f39b631f42 | | model-00067-of-00108.safetensors | Weights | 19.7 GB | a1e01103eb37 | | model-00068-of-00108.safetensors | Weights | 9.7 GB | a3a3c3bf64d6 | | model-00069-of-00108.safetensors | Weights | 19.9 GB | 285ae4895496 | | model-00070-of-00108.safetensors | Weights | 19.7 GB | 6d139332348b | | model-00071-of-00108.safetensors | Weights | 10.0 GB | 7325b5e8b134 | | model-00072-of-00108.safetensors | Weights | 20.0 GB | 1055d776c8fa | | model-00073-of-00108.safetensors | Weights | 19.6 GB | deec9b39cf99 | | model-00074-of-00108.safetensors | Weights | 19.3 GB | 850e48bdcdea | | model-00075-of-00108.safetensors | Weights | 9.7 GB | 8bbe1cf1e260 | | model-00076-of-00108.safetensors | Weights | 19.9 GB | a67af2951e15 | | model-00077-of-00108.safetensors | Weights | 9.7 GB | 1182b941bed8 | | model-00078-of-00108.safetensors | Weights | 19.6 GB | a9dd27406ca2 | | model-00079-of-00108.safetensors | Weights | 19.7 GB | 3db23920d1ba | | model-00080-of-00108.safetensors | Weights | 19.8 GB | aed930ea789f | | model-00081-of-00108.safetensors | Weights | 9.8 GB | 3d7db2959859 | | model-00082-of-00108.safetensors | Weights | 19.8 GB | ff6d6f083495 | | model-00083-of-00108.safetensors | Weights | 19.4 GB | eee163179ec0 | | model-00084-of-00108.safetensors | Weights | 9.8 GB | 51a845d968be | | model-00085-of-00108.safetensors | Weights | 19.4 GB | 08ac68e50849 | | model-00086-of-00108.safetensors | Weights | 19.3 GB | 4925d4ac9f56 | | model-00087-of-00108.safetensors | Weights | 19.8 GB | e9dd5583b90a | | model-00088-of-00108.safetensors | Weights | 19.6 GB | e5e0ba3e6ff8 | | model-00089-of-00108.safetensors | Weights | 12.3 GB | e1990e1a5f87 | | model-00090-of-00108.safetensors | Weights | 10.8 GB | 604a1f065cc5 | | model-00091-of-00108.safetensors | Weights | 19.5 GB | aead1b4e1556 | | model-00092-of-00108.safetensors | Weights | 20.0 GB | 8f675f5752f4 | | model-00093-of-00108.safetensors | Weights | 19.4 GB | 7d7821d1cedb | | model-00094-of-00108.safetensors | Weights | 19.9 GB | c91f417b5a5f | | model-00095-of-00108.safetensors | Weights | 19.5 GB | 8022ceb1cd2d | | model-00096-of-00108.safetensors | Weights | 19.6 GB | 4308a7776edb | | model-00097-of-00108.safetensors | Weights | 19.5 GB | 1248b3c61f9f | | model-00098-of-00108.safetensors | Weights | 19.6 GB | dab1d8104b89 | | model-00099-of-00108.safetensors | Weights | 19.6 GB | 212dcc5b5862 | | model-00100-of-00108.safetensors | Weights | 10.1 GB | 9207b2e3ddda | | model-00101-of-00108.safetensors | Weights | 19.3 GB | ea5c7c6dc75f | | model-00102-of-00108.safetensors | Weights | 19.5 GB | 4f98ac365e35 | | model-00103-of-00108.safetensors | Weights | 9.7 GB | 69104605e7f6 | | model-00104-of-00108.safetensors | Weights | 19.9 GB | 9ffe987ad196 | | model-00105-of-00108.safetensors | Weights | 20.0 GB | 962f0023fbaa | | model-00106-of-00108.safetensors | Weights | 19.5 GB | cfe952b9cf6a | | model-00107-of-00108.safetensors | Weights | 19.4 GB | 33ad10c654ce | | model-00108-of-00108.safetensors | Weights | 9.9 GB | 8ace7b8b1082 | | mtp.safetensors | Weights | 10.5 GB | a0a01bf72ac4 | | .eval_results/aime_2026.yaml | Configuration | 174 B | — | | .eval_results/gpqa.yaml | Configuration | 158 B | — | | .eval_results/hle.yaml | Configuration | 145 B | — | | .eval_results/mmmu_pro.yaml | Configuration | 177 B | — | | .eval_results/swe-bench_pro.yaml | Configuration | 170 B | — | | .eval_results/swe-bench_verified.yaml | Configuration | 184 B | — | | config.json | Configuration | 2.4 KB | — | | model.safetensors.index.json | Configuration | 128.6 KB | — | | processor_config.json | Configuration | 1.1 KB | — | | special_tokens_map.json | Configuration | 517 B | — | | README.md | Documentation | 14.0 KB | — | | chat_template.jinja | Other | 6.3 KB | — | | .gitattributes | Repository | 7.6 KB | — | | tiktoken/tokenizer.model | Tokenizer | 3.6 MB | bc253fd2b702 | | tokenizer.json | Tokenizer | 27.9 MB | 9fb6333a7db8 | | tokenizer_config.json | Tokenizer | 12.1 KB | — | ## License and Download License apache-2.0 Access Open weights, no gate Download size 1.9 TB [Download from Thinking Machines Lab](https://huggingface.co/thinkingmachines/Inkling) Released by Thinking Machines Lab through its official repository on Hugging Face. [Read the license](https://www.apache.org/licenses/LICENSE-2.0). ## 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 diamondComparison conditions not established | 87.2 | [Model Card](https://huggingface.co/thinkingmachines/Inkling) Reported by a third party | Evaluated revision not stated | 2026-07-19 | | MMMU/MMMU_Pro | Task mmmu_pro_standard_10_optionsMetric mmmu_pro_standard_10_optionsComparison conditions not established | 73.5 | [Model Card](https://huggingface.co/thinkingmachines/Inkling) Reported by a third party | Evaluated revision not stated | 2026-07-19 | | MathArena/aime_2026 | Task MathArena/aime_2026Metric MathArena/aime_2026Comparison conditions not established | 97.1 | [Model Card](https://huggingface.co/thinkingmachines/Inkling) Reported by a third party | Evaluated revision not stated | 2026-07-19 | | SWE-bench/SWE-bench_Verified | Task swe_bench_%_resolvedMetric swe_bench_%_resolvedComparison conditions not established | 77.6 | [Model Card](https://huggingface.co/thinkingmachines/Inkling) Reported by a third party | Evaluated revision not stated | 2026-07-19 | | ScaleAI/SWE-bench_Pro | Task SWE_Bench_ProMetric SWE_Bench_ProComparison conditions not established | 54.3 | [Model Card](https://huggingface.co/thinkingmachines/Inkling) Reported by a third party | Evaluated revision not stated | 2026-07-19 | | cais/hle | Task hleMetric hleComparison conditions not established | 46 | [Model Card](https://huggingface.co/thinkingmachines/Inkling) Reported by a third party | Evaluated revision not stated | 2026-07-19 | | harborframework/terminal-bench-2.1 | Task terminalbench_2_1Metric terminalbench_2_1Setup Best Harness. Effort=0.99Comparison conditions not established | 63.8 | [Model Card](https://huggingface.co/thinkingmachines/Inkling) Reported by a third party | Evaluated revision not stated | 2026-07-15 | | joelniklaus/LEXam-hard | Task lexam_hardMetric lexam_hardSetup lighteval, LEXam paper prompts, one response per question, no tools; DeepSeek-R1-0528 judge; mean of the German and English means over the 518 questions, 0-100Comparison conditions not established | 40.82 | [SwissLegalEvals per-sample details (lighteval)](https://huggingface.co/buckets/joelniklaus/SwissLegalEvals) Reported by a third party | Evaluated revision not stated | 2026-07-30 | ## Memory Requirements | Precision | Weights in memory | | --- | --- | | As published | 1.9 TB | | 16-bit | 1904.8 GB | | 8-bit | 952.4 GB | | 4-bit | 476.2 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 | $0.95 / $4.05 | input / output, per million tokens | Oct 12, 2026 | | Together AI | $1.00 / $4.05 | input / output, per million tokens | Oct 12, 2026 | From the [SAVRN Index](https://savrn.com/ai-index/pricing/open-models). ## Questions About Inkling ### How much GPU memory does Inkling need? About 2285.7 GB at 16-bit and 571.4 GB at 4-bit: the weights (952.4B parameters) plus a working margin. A long context needs more. ### What is the cheapest GPU to run Inkling on? At 16-bit, 8x MI355X from $20.72 an hour; at 4-bit, 2x MI355X from $5.18 an hour, at the lowest on-demand prices the SAVRN Index lists. ### Can I use Inkling commercially? Yes. Inkling is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors. ## Similar Models Model · Image and text to text ### [DeepSeek-V4.1-Flash](https://savrn.com/models/deepseek-v4-1-flash) [DeepSeek](https://savrn.com/model-publishers/deepseek-ai) We introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with 552B backbone parameters and support for contexts of up to one million tokens. The model natively processes images and text, and generates text autoregressively. Architecture. DeepSeek-V4.1-Flash adopts a Causal Encoder-Decoder (CED) architecture: a 40-layer Transformer organized as a 20-layer causal encoder followed by a 20-layer decoder. With CED, the decoder's global KV cache is projected from the final encoder hidden states rather than derived from each decoder layer's own hidden states. This allows the model to activate only 8B parameters per token during prefill and 16B during decode, substantially… Open weights mit 763.2B parameters 1,048,576 tokens transformers [View model](https://savrn.com/models/deepseek-v4-1-flash) Model · Image and text to text ### [DeepSeek-V4.1-Flash-UNCENSORED-FP8](https://savrn.com/models/deepseek-v4-1-flash-uncensored-fp8) [SAIFI INDUSTRIES](https://savrn.com/model-publishers/saifiindustries) with permanent weight-level abliteration — the safety guardrails have been surgically removed while preserving MMLU capability, vision, reasoning, MTP (DSpark), and multi-turn coherence. Proprietary weight-level abliteration developed by the dealignai research team. No custom model.py, no runtime hooks, no steering vectors — it's a standard checkpoint that loads exactly like the base model. The refusal circuitry is surgically removed while every capability-critical component (routed experts, Engram memory, CSA2 sparse attention, DSpark draft head, vision tower, router gates, norms, embeddings) is preserved byte-identical to the base. Every response 4-tier graded (HARDREF / SOFTRED / HEDGE /… Open weights mit 763.2B parameters 1,048,576 tokens transformers [View model](https://savrn.com/models/deepseek-v4-1-flash-uncensored-fp8) Model · Image and text to text ### [s](https://savrn.com/models/s) [Lautaro Rodriguez](https://savrn.com/model-publishers/user2simion2) We introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with 552B backbone parameters and support for contexts of up to one million tokens. The model natively processes images and text, and generates text autoregressively. Architecture. DeepSeek-V4.1-Flash adopts a Causal Encoder-Decoder (CED) architecture: a 40-layer Transformer organized as a 20-layer causal encoder followed by a 20-layer decoder. With CED, the decoder's global KV cache is projected from the final encoder hidden states rather than derived from each decoder layer's own hidden states. This allows the model to activate only 8B parameters per token during prefill and 16B during decode, substantially… Access requested at publisher mit 763.2B parameters transformers [View model](https://savrn.com/models/s) Model · Image and text to text ### [Synin-V1.1-Flash](https://savrn.com/models/synin-v1-1-flash) [Synin AI Lab](https://savrn.com/model-publishers/synin) We introduce DeepSeek-V4.1-Flash, a multimodal Mixture-of-Experts (MoE) model with 552B backbone parameters and support for contexts of up to one million tokens. The model natively processes images and text, and generates text autoregressively. Architecture. DeepSeek-V4.1-Flash adopts a Causal Encoder-Decoder (CED) architecture: a 40-layer Transformer organized as a 20-layer causal encoder followed by a 20-layer decoder. With CED, the decoder's global KV cache is projected from the final encoder hidden states rather than derived from each decoder layer's own hidden states. This allows the model to activate only 8B parameters per token during prefill and 16B during decode, substantially… Open weights mit 763.2B parameters 1,048,576 tokens transformers [View model](https://savrn.com/models/synin-v1-1-flash) Model · Image and text to text ### [DeepSeek-V4.1-Flash-Abliterated](https://savrn.com/models/deepseek-v4-1-flash-abliterated) [Alex](https://savrn.com/model-publishers/securepeak) deepseek-ai/DeepSeek-V4.1-Flash with weight-level abliteration of the refusal direction. The refusal direction was computed from 79 harmful vs. 79 benign instruction prompts (per-layer mean-difference of the collapsed residual stream, captured with the official reference implementation, tensor-parallel 4). Exactly 80 tensors were orthogonalized — for each of the 40 backbone layers: - layers.N.attn.wob.weight — attention output projection (writes into the residual stream) - layers.N.ffn.sharedexperts.w2.weight — shared-expert down projection Each weight W was edited as W ← W − r̂ (r̂ᵀ W) with r̂ the unit refusal direction of that layer, removing the model's ability to write the refusal… Open weights mit 756.4B parameters 1,048,576 tokens transformers [View model](https://savrn.com/models/deepseek-v4-1-flash-abliterated) Model · Image and text to text ### [Qwen3.5-397B-A17B-FP8](https://savrn.com/models/qwen3-5-397b-a17b-fp8) [Qwen](https://savrn.com/model-publishers/qwen) Over recent months, we have intensified our focus on developing foundation models that deliver exceptional utility and performance. Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility to empower developers and enterprises with unprecedented capability and efficiency. For more details, please refer to our blog post Qwen3.5. WMT24++: a harder subset of WMT24 after difficulty labeling and rebalancing; we report the averaged scores on 55 languages using XCOMET-XXL. Empty cells (--) indicate scores not yet available or not applicable. MathVision：our model’s score is… Open weights apache-2.0 403.4B parameters 262,144 tokens transformers [View model](https://savrn.com/models/qwen3-5-397b-a17b-fp8) ## Thinking Machines Lab [All models and datasets](https://savrn.com/model-publishers/thinkingmachines) ## Versions - [828496eeae4c](https://savrn.com/models/inkling/versions/828496eeae4c) · current 2026-10-12 ## Explore More - [All image and text to text models](https://savrn.com/models/tasks/image-and-text-to-text) - [All models under apache-2.0](https://savrn.com/models/licenses/apache-2-0) - [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 2026-10-10. - [Hugging Face record](https://huggingface.co/thinkingmachines/Inkling) - [How the hub is built](https://savrn.com/model-hub/methodology)