Sign in

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

Nanbeige4.3-33B vs WeVisOne-VL-2B

WeVisOne-VL-2B and Nanbeige4.3-33B score within a point of each other on agent readiness, 61 (C) and 60.8 (C). Nanbeige4.3-33B leads on reach, risk, agent interface and documentation and upkeep. Both do open-weight models you run yourself.

Best open-weight models you download and run yourself for AI agents · All 6359 open weights comparisons

Nanbeige4.3-33BNanbeige LLM Lab (BOSS Zhipin) · Open-weight model C60.8/100#86 of 160 open-weight models WeVisOne-VL-2BTencent (WeChat Vision) · Open-weight model C61/100#83 of 160 open-weight models

Which one, for what

Nanbeige4.3-33B C

Best for Agents with a 32 GB GPU (INT4) or more, or the 128 GB reference machine at BF16, that want an Apache-tagged 33B model for long coding and tool loops.

Ahead on

  • Reach, 80 against 70
  • Risk, 60 against 55
  • Agent interface, 85 against 70
  • Documentation and upkeep, 56 against 48

Watch for

Independent results so far are Artificial Analysis's alone (Intelligence Index v5.0 20.4, Terminal-Bench 4.0 29.8 per cent); the card's other agentic numbers are Nanbeige's own runs

WeVisOne-VL-2B C

Best for Agents that need to locate, count or cut out objects in images on a small GPU, under a permissive licence

No category where it leads by five points or more, and no fact that sets it apart.

Watch for

The pixel decoder ships as a pickle file

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.

Agent interface Nanbeige4.3-33B +15

Nanbeige4.3-33B 85

The chat template earns 20, Jinja in tokenizer_config.json. The tool-calling format earns 25, with the XML call format in the template and the qwen3_coder parser named for vLLM and SGLang. Trained tool calling earns 20, several calls per turn in the template. Structured output earns 0, as nothing mentions it. Reasoning control earns 10, thinking switched per request with the qwen3 reasoning parser named. Context earns 10 for 512K set in the config with no scaling.

WeVisOne-VL-2B 70

Chat template 20, chat_template.jinja in the repository (5,292 characters), Qwen3-VL's template. Tool-calling format 25, Read for a perception model as what it takes and returns. Trained tool calling 10, Read for a perception model as the inputs it was trained to use. Structured output 15, Read for a perception model as typed outputs. Reasoning control 0, the card says it writes structured output directly with no chain-of-thought reasoning. Context 0, Tencent states no context length for it. 20 + 25 + 10 + 15 + 0 + 0 = 70.

Reach Nanbeige4.3-33B +10

Nanbeige4.3-33B 80

Memory earns 20, since Nanbeige's INT4 build is 22.57 GB, about 27 GB with a fifth added, which fits 32 GB. Runtimes earn 20 from released vLLM, SGLang, Transformers and llama.cpp, which all run the Llama architecture, and Ollama and LM Studio have no build. Builds earn 15 for Nanbeige's INT4, INT8 and FP8 builds of 31 October 2026. Terms earn 15 for Apache 2.0 with nothing added, though only the metadata names it. Download earns 10, with no gate. Hosted earns 0, as no provider or Nanbeige API serves it.

WeVisOne-VL-2B 70

Memory 30, the only build from Tencent is the BF16 safetensors file with its decoder, 4.92 GB, and 4.92 GB × 1.2 = 5.9 GB, within 16 GB. No runtime library has a build. Runtimes 15, the architecture is Qwen3VLForConditionalGeneration, which vLLM 0.33.0, SGLang 0.5.23 and Transformers 5.22.2 run from the safetensors file, three at 5 each. Builds 0, Tencent publishes the BF16 weights only. Terms 15, Apache 2.0 in the card's metadata with the LICENSE file in the repository, and nothing added. Download 10, the Hub API reports gated false on 10 November 2026, so the files download with no account and nothing to accept. Hosted 0, Hugging Face maps no inference provider to it. 30 + 15 + 0 + 15 + 10 + 0 = 70.

Documentation and upkeep Nanbeige4.3-33B +8

Nanbeige4.3-33B 56

The model card earns 11, with intended use, parameters, context and languages but no limitations, architecture or cut-off. Training data earns 0, as nothing describes it. Last change earns 20 for the quantised builds of 31 October 2026. Responsiveness earns 10, a Nanbeige member answering nine of 30 discussions. Version policy earns 5 for a first release, with a config change made and reverted under bare commit titles. File health earns 10, the files and the card agreeing.

WeVisOne-VL-2B 48

Model card 8, Intended use, limitations and the architecture are on the card. Training data 5, A general description in the README of a data engine and over 100 million annotated images, with no sources named and no cut-off. Last change 20, the last commit, on 28 October 2026, put the weights back in BF16, 13 days before this check. Responsiveness 0, the one discussion that isn't a pull request (23 October 2026) has no reply from a tencent member, counted on 10 November by author flags and dates with a script, text not read. Version policy 5, A first release. File health 10, Config, template and card agree on the Qwen3-VL classes, and the card names mask_decoder.pt as required and it's there. 8 + 5 + 20 + 0 + 5 + 10 = 48.

Risk Nanbeige4.3-33B +5

Nanbeige4.3-33B 60

File safety earns 25, safetensors only and no repository code, so the card's trust_remote_code has nothing to run. Provenance earns 15, the lab's own organisation as its arXiv report links it, with a hash for every file. Safety evaluation earns 0 and agent guidance 0, since neither card says anything on either. Licence risk keeps all 15 under Apache 2.0. The security programme earns 5 for BOSS Zhipin's security response centre, which isn't linked to the model.

WeVisOne-VL-2B 55

File safety 15, the backbone is safetensors, but mask_decoder.pt, a PyTorch pickle the card calls a required component, sits beside it. Provenance 15, the files sit in the tencent organisation, which Tencent's GitHub repository for the model links to (10), and the Hub gives a SHA-256 for the large files (5). Safety evaluation 0, no safety results on the card or in the README. Agent guidance 5, General cautions: the card says the release is for research, isn't for autonomous high-impact decisions or identifying individuals, and lists where masks go wrong. Licence risk 15, Apache 2.0 with no use policy, no threshold and no right to change the terms. Security programme 5, Tencent's Security Response Center takes vulnerability reports company-wide, as the Hy3 listing records (5). 15 + 15 + 0 + 5 + 15 + 5 = 55.

Score by category

CategoryWeight this runNanbeige4.3-33BWeVisOne-VL-2BEdge
Capability35%pendingpendingpendingnot scored in this run
Reach25%258070Nanbeige4.3-33B +10
Risk20%206055Nanbeige4.3-33B +5
Agent interface10%108570Nanbeige4.3-33B +15
Documentation and upkeep10%105648Nanbeige4.3-33B +8
Negative events≤1500
TotalC 60.8/100C 61/100

Facts side by side

FactNanbeige4.3-33BWeVisOne-VL-2B
KindOpen-weight modelOpen-weight model
VendorNanbeige LLM Lab (BOSS Zhipin)Tencent (WeChat Vision)
Hosted endpointno (local only)no (local only)
Transports
AuthNoneNone
PricingFreeFree
x402nono
LicenceApache-2.0 (Hub metadata only; the repository has no LICENSE file)Apache-2.0
Read-only variant documentednono
llms.txtnono
Last release2026-10-312026-10-28
Terms last updatedno document linkednot read yet
Privacy policy last updatedno document linkedno document linked

How the scores have moved

Nanbeige4.3-33B

WeVisOne-VL-2B

No score change since it was listed.

Verdicts

Nanbeige4.3-33B

A dense 33B Llama-architecture cousin for long agentic coding and office tasks, with per-request thinking, XML tool calls that stock vLLM and SGLang parse, and 512K of context. Nanbeige's INT4 build needs about 27 GB. Artificial Analysis's first independent runs give it 20.4 on its Intelligence Index, and there are no safety results, training data notes or licence file.

WeVisOne-VL-2B

A small Apache 2.0 perception model that finds, counts, reads and cuts out what you name in an image, and says None when it isn't there. Boxes and points run in vLLM, SGLang or Transformers. Masks and mattes need Tencent's package and a pickle file. Two weeks old, with no questions answered yet and no safety results.

Before you call either

Nanbeige4.3-33B

  1. Serve with vllm serve Nanbeige/Nanbeige4.3-33B --enable-auto-tool-choice --reasoning-parser qwen3 --tool-call-parser qwen3_coder, as the card says, or SGLang with the same two parsers
  2. On a 32 GB GPU use Nanbeige/Nanbeige4.3-33B-INT4 with the same vLLM or SGLang flags; the INT8 and FP8 builds need about 43 GB
  3. Keep tool_call_format at its default xml, since qwen3_coder parses that format; the json option writes Hermes-style JSON calls, which need a different parser such as vLLM's hermes
  4. Pass chat_template_kwargs {"preserve_thinking": true} for multi-turn tool use and coding, and {"enable_thinking": false} to skip reasoning, as the card recommends
  5. Use temperature 1.0, top_p 0.95 and up to 65,536 new tokens for agent work, and temperature 0.6 for reasoning and chat, per the card
  6. Send your own system prompt, since without one the template injects a Chinese identity line, or a Chinese tool-expert prompt when tools are passed
  7. Send one system message at the start only, as the template raises an error for a later one
  8. In plain Transformers pass use_cache=True to generate, since config.json and generation_config.json set it false; trust_remote_code isn't needed

WeVisOne-VL-2B

  1. Use Tencent's package: WeVisOneVL("tencent/WeVisOne-VL-2B", backend="vllm") then model.predict(image, task=..., query=...)
  2. Tasks are detection, referring, point, count_box, count_point, ocr, ocr_grounding, segmentation and matting
  3. A None answer means the target isn't in the image. Treat it as a result, not an error
  4. mask_decoder.pt is a pickle. Load it only from the tencent repository, and check its SHA-256 against the Hub's
  5. Send one image a turn. Video grids are rejected by the decoder
  6. The card says it isn't for identifying individuals or for high-impact decisions on its own

Questions

Which is better for AI agents, Nanbeige4.3-33B or WeVisOne-VL-2B?

WeVisOne-VL-2B and Nanbeige4.3-33B score within a point of each other on agent readiness, 61 (C) and 60.8 (C). Nanbeige4.3-33B leads on reach, risk, agent interface and documentation and upkeep.

Other comparisons with Nanbeige4.3-33B or WeVisOne-VL-2B

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