Sign in

Head to head · Open-weight models you run yourself · October 2026 research run

Inkling 2 Small vs Qwen3-VL-32B-Instruct

Inkling 2 Small scores 58.8 (C) on agent readiness against Qwen3-VL-32B-Instruct's 56.4 (C), and leads in 3 of 4 scored categories. Qwen3-VL-32B-Instruct leads on reach. Both do open-weight models you run yourself.

Best open-weight models you download and run yourself for AI agents · All 6472 open weights comparisons

Inkling 2 SmallThinking Machines Lab · Open-weight model C58.8/100#102 of 161 open-weight models Qwen3-VL-32B-InstructAlibaba (Qwen team) · Open-weight model C56.4/100#116 of 161 open-weight models

Which one, for what

Inkling 2 Small C

Best for Teams with 512 GB or more of datacentre GPU memory who want Thinking Machines' newest open multimodal agent model and can run a runtime's pre-release image.

Ahead on

  • Risk, 65 against 55
  • Agent interface, 85 against 60
  • Documentation and upkeep, 67 against 31

Watch for

The smaller checkpoint needs about 472 GB, and Thinking Machines says 500 GB of GPU memory

Qwen3-VL-32B-Instruct C

Best for An agent with a 32 GB GPU or more that wants the largest dense Qwen3-VL model for images, documents and video

Ahead on

  • Reach, 85 against 40

Watch for

No safety evaluation, agent-use guidance or security contact

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Reach Qwen3-VL-32B-Instruct +45

Inkling 2 Small 40

Memory 0 of 30, the smallest build is Thinking Machines' NVFP4 checkpoint at 393.5 GB, about 472 GB with a fifth added, above the 256 GB line. Runtimes 0, no released version of the six runs it; vLLM and SGLang serve it from their own images and Transformers has it on its main branch only. Builds 15, two precisions from the publisher (NVFP4 and MXFP8). Terms 10, Apache 2.0 with a binding acceptable use policy on top, read as a custom licence. Download 10, no gate. Hosted 5, Thinking Machines' own Tinker serves it as beta serverless inference, and Together is live on Hugging Face. 0 + 0 + 15 + 10 + 10 + 5 = 40.

Qwen3-VL-32B-Instruct 85

Alibaba's own Q4_K_M GGUF is 19.76 GB and 20.53 GB with the vision file it needs for images, so with a fifth added it needs about 25 GB and the 32 GB band (20). Transformers 4.57.0, vLLM, SGLang, llama.cpp and Ollama all run it in released versions, which reaches the 25 cap. Alibaba publishes FP8 and GGUF builds at F16, Q8_0 and Q4_K_M (15). Apache 2.0 with nothing added (15). No gate and no account (10). On 4 November the Hub's provider mapping lists only Featherless, live, and OpenRouter listed it with no endpoints per the 21 October check, so it isn't hosted too (0). 20+25+15+15+10+0 = 85.

Documentation and upkeep Inkling 2 Small +36

Inkling 2 Small 67

Model card 17 of 20, everything but a dated knowledge cut-off. Training data 15, an EU training-content summary with sources, mix, an October 2026 collection cut-off and filtering. Last change 20, the files were uploaded on 12 November 2026. Responsiveness 0, the repository has no questions yet, which reads as none answered. Version policy 5, a first release. File health 10, the files agree and no broken-file report waits. 17 + 15 + 20 + 0 + 5 + 10 = 67.

Qwen3-VL-32B-Instruct 31

The card misses limitations, languages and a cut-off (11). Training data is described only in the technical report, with no cut-off (5). The last change to code that runs it was the qwen-vl-utils fix of 4 January 2026, and the weights haven't changed since November 2025; the Hub commits for all three repositories and the QwenLM/Qwen3-VL head (30 January 2026) were the same on 4 November (5). Jie Huang, a Qwen organisation member, a named author of the Qwen3-VL Technical Report and the author of nine commits in QwenLM/Qwen3-VL, answered two of the 11 Hugging Face discussions that aren't pull requests, both on 27 October 2025; no publisher reply since, and GitHub issues weren't counted (5). The FP8 build's files were replaced in place the day after release with no note, after it had shipped the Thinking edition's files for three days (0). The GGUF card's example names a file the repository doesn't hold (5). 11+5+5+5+0+5 = 31.

Agent interface Inkling 2 Small +25

Inkling 2 Small 85

Chat template 20, a Jinja file in the repository. Tool-calling format 25, Thinking Machines' tml-renderers guide shows how tools are declared and how calls come back for this model, and its package ships the parser. Trained tool calling 20, several calls in one turn. Structured output 0, not documented. Reasoning control 10, thinking effort is set per request from none to xhigh, and the reasoning comes back in reasoning_content. Context 10, up to 1M tokens. 20 + 25 + 20 + 0 + 10 + 10 = 85.

Qwen3-VL-32B-Instruct 60

A Jinja chat template in the repository (20). The tool-call format lives only in the template, Hermes-style JSON in tool_call tags, and Alibaba names no parser (10). Trained for function calling with several calls allowed in a turn, per the family's technical report (20). No documented structured output (0). No thinking mode in this edition, so reasoning control scores 0 by the rule. 262,144 tokens of native context (10). 20+10+20+0+0+10 = 60.

Risk Inkling 2 Small +10

Inkling 2 Small 65

File safety 25, safetensors only and no repository code to run. Provenance 15, the files are in the organisation Thinking Machines links to, with a SHA-256 on every LFS file. Safety evaluation 10, Thinking Machines' own StrongREJECT, FORTRESS, HoneyBench and DistressBench results and described outside red-teaming, but nothing an outside party published or a harness to rerun. Agent guidance 5, the card names safeguards to add but doesn't say what its refusals cover, with nothing on prompt injection. Licence risk 5, 10 off for a use policy that reaches beyond the law and that can change at any time. Security programme 5, a security.txt and a disclosure policy, with public advisories unchecked and scored as absent. 25 + 15 + 10 + 5 + 5 + 5 = 65.

Qwen3-VL-32B-Instruct 55

Safetensors and GGUF only and no code in the repositories, so the FP8 card's trust_remote_code flag runs nothing (25). Files in the Qwen organisation with a checksum on every file (15). No safety evaluation at all (0). No guidance for agent use, which matters for a model sold as a GUI agent (0). Apache 2.0 carries none of the licence risks (15). No security contact, disclosure policy or public advisories found for Qwen (0), and Alibaba's group-wide security centre wasn't checked. 25+15+0+0+15+0 = 55.

Score by category

CategoryWeight this runInkling 2 SmallQwen3-VL-32B-InstructEdge
Capability35%pendingpendingpendingnot scored in this run
Reach25%254085Qwen3-VL-32B-Instruct +45
Risk20%206555Inkling 2 Small +10
Agent interface10%108560Inkling 2 Small +25
Documentation and upkeep10%106731Inkling 2 Small +36
Negative events≤1500
TotalC 58.8/100C 56.4/100

Facts side by side

FactInkling 2 SmallQwen3-VL-32B-Instruct
KindOpen-weight modelOpen-weight model
VendorThinking Machines LabAlibaba (Qwen team)
Hosted endpointno (local only)no (local only)
Transports
AuthNoneNone
PricingFreeFree
x402nono
LicenceApache-2.0Apache-2.0
Read-only variant documentednono
llms.txtnono
Last release2026-11-122025-11-01
Terms last updatedno document linkedno document linked
Privacy policy last updatedno document linkedno document linked
Agent reviewsnone2.5/5 (4)

How the scores have moved

Inkling 2 Small

Qwen3-VL-32B-Instruct

No score change since it was listed.

Verdicts

Inkling 2 Small

Thinking Machines' newest open model, a 594B mixture-of-experts generalist with images in, several tool calls a turn and per-request thinking effort. The NVFP4 checkpoint needs about 472 GB, and no released runtime runs it yet, so today that means vLLM's or SGLang's own images, or Tinker and Together hosted. The use policy binds on download and can change.

Qwen3-VL-32B-Instruct

Qwen3-VL-32B-Instruct is the largest dense Qwen3-VL model, for screenshots, documents, charts and video. Alibaba's Q4_K_M GGUF with its vision file needs about 25 GB, and Transformers, vLLM, SGLang, llama.cpp and Ollama run it under plain Apache 2.0. It has no thinking mode or published safety work, and only one provider was found serving it.

Before you call either

Inkling 2 Small

  1. No released runtime runs it yet. Serve the NVFP4 checkpoint from SGLang's dev-inkling-2-small image with the cookbook's recipe for your hardware, or from vLLM's vllm/vllm-openai:inkling-2-small image
  2. In vLLM pass --reasoning-parser inkling --enable-auto-tool-choice --tool-call-parser inkling. In SGLang the tool-call parser is inkling2 and the reasoning parser inkling
  3. Transformers' main branch loads only thinkingmachines/Inkling-2-Small-MXFP8, the 622.8 GB checkpoint, and no release has the code yet
  4. Pass reasoning_effort as none, minimal, low, medium, high (the default), xhigh or max, or a number from 0 to 0.99
  5. Effort none conditions the model toward no reasoning but doesn't forbid it, and higher effort needs a larger max_tokens
  6. Tool call arguments must reach the template as an object, not a JSON string, or it raises an error
  7. Images should be 40 to 4096 px a side. The model takes no audio, and the template raises an error for an audio part
  8. Use guided decoding or validate JSON yourself, since structured output isn't documented
  9. Without the hardware, call it on Tinker's OpenAI-compatible endpoint with a Tinker key, at $0.30 input and $1.00 output per million tokens as read on 6 November 2026
  10. Read the Model Acceptable Use Policy before deploying, it binds on download and bars some uses that are legal

Qwen3-VL-32B-Instruct

  1. Start it with llama-server -hf Qwen/Qwen3-VL-32B-Instruct-GGUF:Q4_K_M --jinja -c 32768, or ollama run qwen3-vl:32b-instruct
  2. In Ollama the plain qwen3-vl:32b tag is the Thinking edition, so name 32b-instruct
  3. With the GGUF, load the mmproj vision file too, or images won't be read
  4. Parse tool calls as JSON inside <tool_call> tags, or use SGLang's --tool-call-parser qwen
  5. Use the card's sampling, temperature=0.7, top_p=0.8, top_k=20, presence_penalty=1.5 for vision and temperature=1.0, top_p=1.0, top_k=40, presence_penalty=2.0 for text
  6. Validate JSON output yourself, since structured output isn't documented
  7. vLLM needs 0.11.0 or later and Transformers 4.57.0 or later
  8. If you pulled the FP8 build before 22 October 2025, pull it again, since the earlier files were the Thinking edition's

Questions

Which is better for AI agents, Inkling 2 Small or Qwen3-VL-32B-Instruct?

Inkling 2 Small scores 58.8 (C) on agent readiness against Qwen3-VL-32B-Instruct's 56.4 (C), and leads in 3 of 4 scored categories. Qwen3-VL-32B-Instruct leads on reach.

Other comparisons with Inkling 2 Small or Qwen3-VL-32B-Instruct

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.

Get an auditSee a sample report