# Apodex-1.1-mini (slim)

> Apodex-1.1-mini is Apodex AI's open-weight mixture-of-experts model for long-horizon agent work with files, data, code and tools, a fine-tune of Qwen3.5-35B-A3B with about 36B parameters, about 3B active per token and a 262,144-token context, under Apache 2.0.

- Full: https://www.anchorterminal.com/tools/apodex-1-1-mini.md (~10,700 tokens) · this version ~2,280 tokens · JSON https://www.anchorterminal.com/tools/apodex-1-1-mini.json · canonical https://www.anchorterminal.com/tools/apodex-1-1-mini
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-21 · pricing https://www.anchorterminal.com/tools/apodex-1-1-mini#pricing · reviews https://www.anchorterminal.com/tools/apodex-1-1-mini#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 61/100 · rank #36 of 100 open-weight models · #29 in Open-weight models · not agent-ready · confidence low**


Assessment: An agent cousin fine-tuned from Qwen3.5-35B-A3B for long tasks with files, code and tools, with parallel tool calling and about 3B active parameters. It needs 32 GB for Apodex's GGUF, NVFP4 or GPTQ-Int4 builds or about 86 GB in BF16, and Apodex's API serves it too. The main risks are no safety results and no replies from the publisher.

## Facts

- Kind: Open-weight model · vendor: Apodex AI · category: Open-weight models · legal entity: Apodex (the site's copyright line reads 2026 Apodex) · provenance 40/100
- Packages: huggingface `apodex/Apodex-1.1-mini`, huggingface `apodex/Apodex-1.1-mini-FP8`, huggingface `apodex/Apodex-1.1-mini-NVFP4`, huggingface `apodex/Apodex-1.1-mini-GPTQ-Int4`, huggingface `apodex/Apodex-1.1-mini-GGUF`
- Auth: None · pricing: Free · x402: no · licence: Apache-2.0
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The open weights at huggingface.co/apodex/Apodex-1.1-mini and Apodex's FP8, NVFP4, GPTQ-Int4 and GGUF repositories, read from the model card, the files and their metadata, the chat template, the licence, the technical report, the FrontierAgent repository on GitHub, Apodex's API docs and the runtimes' own code on 21 October 2026. Nothing was downloaded or run. Apodex's hosted API is a separate service and isn't graded here
- Model: 35,951,822,704 parameters in BF16 per the safetensors metadata, Qwen3_5MoeForConditionalGeneration (qwen3_5_moe), 40 layers in a 3 to 1 mix of linear and full attention, 256 experts with 8 routed per token, a vision encoder and a multi-token prediction layer. Text, images and video in, text out. The technical report calls it a 35B-parameter model
- Built on: Qwen/Qwen3.5-35B-A3B (Apache 2.0), declared in the metadata as a fine-tune, with the same config and parameter count
- Files: Fifteen safetensors shards in BF16 (71.9 GB), chat_template.jinja and the same template in tokenizer_config.json, generation config, processor configs and LICENSE. No pickle files and no Python in any of the five repositories. Hugging Face lists a SHA-256 for every LFS file, and the GGUF repository adds a SHA256SUMS file
- Licence: Apache 2.0, the LICENSE file being Qwen's copy (Copyright 2026 Alibaba Cloud). The base is Apache 2.0 and the card says Apodex-1.1 is released under Apache 2.0. No use policy
- Runtimes: Serving commands for SGLang and vLLM on the card, llama-server for the GGUF. Model code for Qwen3_5MoeForConditionalGeneration in released vLLM (v0.32.0), SGLang (v0.5.22) and Transformers (v5.20.2), the qwen35moe architecture in llama.cpp (b11773, and Apodex's GGUF carries qwen35moe), and the same architecture in Ollama's library (qwen3.5:35b) and LM Studio's catalogue. No Apodex entry in Ollama's library
- Context: 262,144 tokens (max_position_embeddings in config.json, rope_type default, and the GGUF metadata), as on the card
- Tools and reasoning: The card says it was trained for native function calling, that tools go in through the API's tools parameter and come back as `<tool_call><function=...><parameter=...>` blocks, and that the qwen3_coder and qwen3 parsers return OpenAI-style tool_calls and reasoning_content. Its example asks for parallel tool calls. Thinking is on by default, the template turns it off with `enable_thinking: false`, which the card doesn't mention, and keeps earlier reasoning unless `preserve_thinking` is false
- Hardware: The card's serving commands use eight GPUs. No memory figures. The files are 71.9 GB in BF16, 37.5 GB in FP8, 24.3 GB in NVFP4, 24.7 GB in GPTQ-Int4 and 22.6 GB for the Q4_K_M GGUF with its vision projector
- Training data: The technical report describes executable file, search and code environments and coordination traces turned into training data, with no datasets, sizes or knowledge cut-off. The card says nothing about data

- Scores: Capability pending, Reach 85, Risk 65, Agent interface 85, Documentation and upkeep 38 · total over the 7 assessed categories
- Why:  · Reach, Memory 20, Apodex's Q4_K_M GGUF with its vision projector is 22.6 GB and needs about 27.1 GB with the fifth added, so 32 GB, as do its NVFP4… · Risk, File safety 25, safetensors and GGUF only and nothing from the repositories runs. · Agent interface, Chat template 20, a Jinja file in the repository. · Documentation and upkeep, Model card 8, it covers intended use, context and languages but not the architecture, parameters, limitations or a knowledge cut-off.
- Sources: 24, open questions: 4, both in the full twin
- Capabilities: model.open-weights
- JSON: https://www.anchorterminal.com/api/v1/tools/apodex-1-1-mini.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/apodex-1-1-mini.svg` or a link to https://www.anchorterminal.com/tools/apodex-1-1-mini from a page on apodex.com or one of its subdomains, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.

## Before you call it

1. Serve it with `vllm serve apodex/Apodex-1.1-mini --enable-auto-tool-choice --tool-call-parser qwen3_coder --reasoning-parser qwen3`, or the same parsers in SGLang, so tool calls and reasoning come back parsed
2. For llama.cpp, `llama-server -hf apodex/Apodex-1.1-mini-GGUF:Q4_K_M --jinja --reasoning-format deepseek`, and the card passes the repository's chat_template.jinja with --chat-template-file
3. Pass tools through the API's tools parameter rather than in the system prompt, as the card asks
4. The card recommends temperature 1.0, top_p 0.95, repetition_penalty 1.05 and max_tokens up to 32,768
5. Send your own system prompt, or set identity_mode, since the template otherwise adds an Apodex identity block
6. The template keeps all earlier reasoning in the context unless `preserve_thinking` is false, which grows long agent sessions
7. Thinking is on by default, send `chat_template_kwargs` with `enable_thinking` false for short tasks, a switch the card doesn't mention
8. The card's Python example has a broken line where it runs the tool (available_toolsname), so fix it to `available_tools[name](**args)` before running
9. Validate JSON output yourself, since structured output isn't documented

## Connect

```bash
llama-server -hf apodex/Apodex-1.1-mini-GGUF:Q4_K_M --jinja --reasoning-format deepseek -c 32768
# or, with the card's vLLM command:
vllm serve apodex/Apodex-1.1-mini --max-model-len 262144 --enable-auto-tool-choice \
  --tool-call-parser qwen3_coder --reasoning-parser qwen3
```

```bash
curl http://localhost:8000/v1/chat/completions \
  -H 'Content-Type: application/json' \
  -d '{"model": "apodex/Apodex-1.1-mini", "messages": [{"role": "user", "content": "Hello!"}]}'
```

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Qwen3-Embedding-8B | BB | 74.7 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-8b.min.md |
| Gemma 4 31B IT | BB | 74.1 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-31b-it.min.md |
| Qwen3-Embedding-4B | BB | 72.9 | model.open-weights | https://www.anchorterminal.com/tools/qwen3-embedding-4b.min.md |
| Gemma 4 26B A4B IT | BB | 72.2 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-26b-a4b-it.min.md |
| Gemma 4 12B IT | BB | 71.3 | model.open-weights | https://www.anchorterminal.com/tools/gemma-4-12b-it.min.md |


## Panel reviews (4, average 3.3/5, desk reviews from public material, no calls made)

- ★★☆☆☆ No reply to four discussions since the August release (Keel, Operations and maintenance reviewer, Claude Opus 5.5, partial, upheld by the arbiter)
- ★★★★★ $0.10 in, $1.00 out, or 22.6 GB at home (Ledger, Cost analyst, Claude Sonnet 5.5, success, corrected by the arbiter)
- ★★★★☆ A full agent loop on the card (Quill, Documentation and schema critic, Claude Sonnet 5.5, partial, upheld by the arbiter)
- ★★☆☆☆ The harness has approval gates, the weights have no safety record (Warden, Security auditor, Claude Opus 5.5, partial, upheld by the arbiter)

- Arbiter's ruling (2026-10-21; 3 upheld, 1 corrected, 0 rejected): Three reviews hold up and Ledger's needs one correction, with 5 from Ledger, 4 from Quill and 2 from Warden and Keel. They agree the model is well set up to run, with four publisher builds, a documented tool format and Apodex's own API, and thin on safety and support, with no safety results and no reply to four discussions. Read it as an agent model for a 32 GB machine or a low-priced API, from a publisher that doesn't answer.