Start with the workload. Then choose the machine.
Compare model profiles and memory requirements before you commit to hardware. Each guide separates what the catalog says from what still needs testing.
Refine results
Memory filters match a recorded catalog threshold in at least one runtime profile. An unspecified threshold never counts as zero. This does not verify one-card fit or measured memory use. Sources are reviewed; no B3IQ benchmark is published for this cohort.
12 of 12 model guides
How to read the evidence
- Source reviewed
- Publisher identity, access and license.
- Catalog configured
- Recorded runtime and artifact settings.
- Planning figure
- Arithmetic or guidance with stated exclusions.
- B3IQ measured
- A reproducible workload record. None published for these guides.
- Ready on your machine
- Installation and capacity need confirmation.
Chat / reasoning
Llama 3.1 8B
A text assistant checkpoint for chat, summarization and retrieval-backed answers. Start here when you want a smaller Llama deployment to evaluate with your own prompts.
- Model creator
- Meta
- Catalog GPU threshold
- 8 GB
Chat / reasoning
Llama 3.3 70B
Meta's instruction-tuned text model for multilingual dialogue. Evaluate it when a smaller assistant misses the quality bar and you can budget more GPU memory.
- Model creator
- Meta
- Catalog GPU threshold
- 80 GB
Chat / reasoning
Qwen3 8B
A compact Qwen text model with thinking and non-thinking modes upstream. It is a useful starting point for comparing reasoning behavior with a modest weight footprint.
- Model creator
- Qwen / Alibaba Cloud
- Catalog GPU threshold
- Not specified
Chat / reasoning
Qwen3 14B
A Qwen text model between the smaller and larger dense checkpoints. Compare it on your own assistant or coding prompts before committing to a larger machine.
- Model creator
- Qwen / Alibaba Cloud
- Catalog GPU threshold
- 16–24 GB across recorded profiles; other profiles unspecified
Chat / reasoning
Qwen3 32B
A larger dense Qwen model for text generation and reasoning. It gives you another quality point to test before moving to a much larger checkpoint.
- Model creator
- Qwen / Alibaba Cloud
- Catalog GPU threshold
- 32 GB
Chat / reasoning
Qwen3 Coder 30B
A mixture-of-experts checkpoint built for coding and tool use. Evaluate it against repository tasks with checkable outcomes, such as passing tests or a correct patch.
- Model creator
- Qwen / Alibaba Cloud
- Catalog GPU threshold
- 32 GB
Chat / reasoning
gpt-oss 20B
OpenAI's smaller open-weight reasoning model, with tool use and structured output support documented upstream. Consider it for a private reasoning or agent evaluation.
- Model creator
- OpenAI
- Catalog GPU threshold
- 24 GB
Chat / reasoning
gpt-oss 120B
The larger gpt-oss reasoning checkpoint. Evaluate it when task quality warrants a larger memory budget, especially for tool-using workflows with verifiable results.
- Model creator
- OpenAI
- Catalog GPU threshold
- 80 GB
Chat / reasoning
Gemma 3 27B
Google's instruction-tuned Gemma checkpoint supports text and image understanding upstream. The current catalog profile is text chat; image serving needs separate validation.
- Model creator
- Google DeepMind
- Catalog GPU threshold
- 24 GB
Chat / reasoning
Qwen2.5-VL 7B
A vision-language model for questions about images and visual documents. Use your real image types when assessing extraction accuracy and grounding.
- Model creator
- Qwen / Alibaba Cloud
- Catalog GPU threshold
- 7 GB
Embeddings
BGE-M3
An embedding model for retrieval. It turns documents and queries into searchable representations; pair it with a text model when building retrieval-backed answers.
- Model creator
- BAAI
- Catalog GPU threshold
- Not specified
Speech to text
Whisper large-v3
A speech recognition checkpoint for transcription and speech translation. It serves an audio workflow rather than a text chat endpoint.
- Model creator
- OpenAI
- Catalog GPU threshold
- 6 GB; other profiles unspecified
A guide is the start of an evaluation.
These pages cover reviewed upstream models and catalog profiles. They do not promise live availability or a tested deployment. Use the guide to scope your workload, then confirm a configuration with us.
Read the evidence standard →
