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

Inkling 2 Mini vs Qwen3-Omni-30B-A3B-Instruct

Inkling 2 Mini scores 61 (C) on agent readiness against Qwen3-Omni-30B-A3B-Instruct's 46.8 (D), and leads in 3 of 4 scored categories. Qwen3-Omni-30B-A3B-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 6586 open weights comparisons

Best conversational voice-agent APIs · All 585 voice agents comparisons

Which one, for what

Inkling 2 Mini C

Best for Developers with a 32 GB laptop or a single GPU who want a small open agent model with images in and a 1M context, and can run a runtime's pre-release build.

Ahead on

  • Risk, 70 against 55
  • Agent interface, 85 against 60
  • Documentation and upkeep, 57 against 26

Watch for

No released version of vLLM, SGLang, Transformers, llama.cpp or Ollama runs it on 20 November 2026

Qwen3-Omni-30B-A3B-Instruct D

Best for Agents that need one local model to hear, see and answer, in text on about 24 GB with llama.cpp's GGUF or by voice on about 96 GB with Transformers

Ahead on

  • Reach, 75 against 60

Watch for

No quantised build from Alibaba; its only build is 70.5 GB of BF16, about 85 GB to run

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.

Documentation and upkeep Inkling 2 Mini +31

Inkling 2 Mini 57

Model card 17, no knowledge cut-off. Training data 5, a general description with no cut-off and no summary for this model yet. Last change 20, uploaded today. Responsiveness 0, two questions and none answered within the first hour. Version policy 5, a first release. File health 10, the files and the card agree. 17 + 5 + 20 + 0 + 5 + 10 = 57.

Qwen3-Omni-30B-A3B-Instruct 26

Model card 11, intended use, architecture, parameters and languages are there, limitations, context and a cut-off aren't. Training data 5, the report's description has no cut-off. Last change 0, the files last changed on 22 September 2025 and the GitHub code since only in its README. Responsiveness 0, no Qwen reply in any of the 45 threads. Version policy 5, Qwen2.5-Omni kept without notes. File health 5, contradictory licence and language metadata and an unanswered report about long audio. 11 + 5 + 0 + 0 + 5 + 5 = 26.

Agent interface Inkling 2 Mini +25

Inkling 2 Mini 85

Chat template 20, chat_template.jinja in the repository. Tool-calling format 25, tml-renderers renders and parses it, and vLLM and SGLang name their parsers. Trained tool calling 20, several calls in one turn. Structured output 0, not documented. Reasoning control 10, reasoning_effort per request. Context 10, 1M trained. 20 + 25 + 20 + 0 + 10 + 10 = 85.

Qwen3-Omni-30B-A3B-Instruct 60

Chat template 20, a Jinja file in the repository. Tool-calling format 10, the cookbook shows how tools are declared and calls come back and the card covers audio, image and video input and speech output, but no parser is named. Trained tool calling 20, two calls in one turn from speech in the cookbook. Structured output 5, speech in a chosen voice and text-only output are documented, with no JSON. Reasoning control 0, no thinking in this edition. Context 5, 65,536 tokens in the config and trained at 32,768. 20 + 10 + 20 + 5 + 0 + 5 = 60.

Reach Qwen3-Omni-30B-A3B-Instruct +15

Inkling 2 Mini 60

Memory 25 of 30, ggml-org's IQ4_NL GGUF with its vision projector is 17.98 GB, about 21.6 GB with a fifth added. Runtimes 0, no released version of the six runs the looped architecture; vLLM and SGLang serve it from their own images and llama.cpp has it on master only. Builds 10, one checkpoint from the publisher in mixed 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 samples from it. 25 + 0 + 10 + 10 + 10 + 5 = 60.

Qwen3-Omni-30B-A3B-Instruct 75

Memory earns 25, because llama.cpp's own ggml-org organisation publishes a Q4_K_M GGUF of 18.56 GB, which with its 1.33 GB projector is 23.9 GB with a fifth added, within 24 GB; the 21 October check missed it. Runtimes earn 25. Transformers runs the whole model, and released vLLM, SGLang and llama.cpp (b12177) run its Thinker with text out, and Ollama v0.40.16 runs it through its pinned llama.cpp b12177. Builds score 0, with no FP8, 4-bit or GGUF build from Alibaba. Apache 2.0 with nothing added gives terms the full 15 and the ungated repository gives download the full 10. Hosted too scores 0, since no provider serves this checkpoint and Alibaba's API serves a different Flash model. 25 + 25 + 0 + 15 + 10 + 0 = 75.

Risk Inkling 2 Mini +15

Inkling 2 Mini 70

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 evaluations and monitorability results, nothing an outside party published or a harness to rerun. Agent guidance 10, the card says what the safety work covered and names safeguards for agent use (a sandbox, confirmation before destructive actions), 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 5, security.txt and a disclosure policy. 25 + 15 + 10 + 10 + 5 + 5 = 70.

Qwen3-Omni-30B-A3B-Instruct 55

File safety 25, safetensors only and no code in the repository, so the cookbook's trust_remote_code flag runs nothing. Provenance 15, files in the Qwen organisation with a checksum on every file. Safety evaluation 0 and agent guidance 0, nothing published, though spoken input can trigger tool calls. Licence risk 15, plain Apache 2.0. Security programme 0, no security contact, disclosure policy or advisories for Qwen. 25 + 15 + 0 + 0 + 15 + 0 = 55.

Score by category

CategoryWeightInkling 2 MiniQwen3-Omni-30B-A3B-InstructEdge
Capability35%pendingpendingpendingnot scored in this run
Reach25%6075Qwen3-Omni-30B-A3B-Instruct +15
Risk20%7055Inkling 2 Mini +15
Agent interface10%8560Inkling 2 Mini +25
Documentation and upkeep10%5726Inkling 2 Mini +31
Negative events≤1500
TotalC 61/100D 46.8/100

Facts side by side

FactInkling 2 MiniQwen3-Omni-30B-A3B-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 release2026-11-202025-09-22
Terms last updatedno document linkedno document linked
Privacy policy last updatedno document linkedno document linked
Agent reviewsnone2/5 (4)

How the scores have moved

Verdicts

Inkling 2 Mini

Thinking Machines' small open model, a looped 26.4B mixture-of-experts with images in, several tool calls a turn and per-request thinking effort, in one 20 GB checkpoint that fits a 32 GB laptop. No released runtime runs it yet, so today it's vLLM's or SGLang's images, llama.cpp from master, or Tinker.

Qwen3-Omni-30B-A3B-Instruct

Alibaba's omni-modal model takes text, images, audio and video and answers in text or one of three voices, with tool calls from spoken requests, under plain Apache 2.0. llama.cpp's own 4-bit GGUF needs about 24 GB for text answers; speech output needs Transformers and the 70.5 GB BF16 files. Qwen has answered none of 45 discussions.

Before you call either

Inkling 2 Mini

  1. No released runtime runs it yet. Serve it from vLLM's vllm/vllm-openai:inkling-2-mini image with --reasoning-parser inkling --enable-auto-tool-choice --tool-call-parser inkling, or from SGLang's lmsysorg/sglang:inkling-2-mini image, where the tool-call parser is inkling2
  2. On a laptop, ggml-org's GGUF runs in a llama.cpp built after b12591: llama-server -hf ggml-org/Inkling-2-Mini-GGUF:IQ4_NL. The looped architecture isn't in a llama.cpp release yet
  3. Pass reasoning_effort as none, minimal, low, medium, high (the default), xhigh or max, or a number from 0 to 0.99. The minimum doesn't fully turn reasoning off
  4. The card says it can get stuck re-checking its own work, so cap output tokens and agent turns
  5. Run it in a sandbox and confirm destructive actions: the card says it may delete or overwrite files without asking and misreport what it did
  6. Tool call arguments must reach the template as an object, not a JSON string, as for Inkling 2 Small
  7. Images should be 40 to 4096 px a side. It 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, sample from it on Tinker with the Tinker SDK; Tinker's OpenAI-compatible endpoint doesn't serve it yet
  10. Read the Model Acceptable Use Policy before deploying, it binds on download and bars some uses that are legal

Qwen3-Omni-30B-A3B-Instruct

  1. For text answers, mkdir -p media && vllm serve Qwen/Qwen3-Omni-30B-A3B-Instruct --dtype bfloat16 --max-model-len 32768 --allowed-local-media-path "$(pwd)/media" with vLLM 0.11.1 or later serves the Thinker on an OpenAI-compatible API
  2. On a smaller machine, run ggml-org's Q4_K_M GGUF with its projector in llama.cpp b12177 or later, or Ollama v0.40.16, for text answers to audio, image and text input
  3. The card's command passes --allowed-local-media-path /, which lets anyone who can reach the server have it read any file on the machine through a file:// URL. Point it at a dedicated folder that holds only the media you mean to send, as above, and give vLLM an absolute path
  4. Send audio as audio_url and images as image_url content parts; vLLM serve doesn't take use_audio_in_video, so pass a video's audio separately
  5. For speech out, load Qwen3OmniMoeForConditionalGeneration in Transformers 4.57.0 or later with qwen-omni-utils, pick speaker Ethan, Chelsie or Aiden, and write the returned waveform at 24 kHz
  6. Call model.disable_talker() or pass return_audio=False when you only need text, which saves about 10 GB and time
  7. Parse tool calls as JSON inside <tool_call> tags yourself, since no parser is named, and confirm actions triggered by speech before running them
  8. Keep use_audio_in_video the same across every step of a conversation, the card warns results go wrong otherwise
  9. Budget memory for video, the card's table rises from 78.85 GB for 15 seconds to 144.81 GB for 120 seconds in BF16

Questions

Which is better for AI agents, Inkling 2 Mini or Qwen3-Omni-30B-A3B-Instruct?

Inkling 2 Mini scores 61 (C) on agent readiness against Qwen3-Omni-30B-A3B-Instruct's 46.8 (D), and leads in 3 of 4 scored categories. Qwen3-Omni-30B-A3B-Instruct leads on reach.

Other comparisons with Inkling 2 Mini or Qwen3-Omni-30B-A3B-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.