CHOOSE A PRODUCT · BUILD WITH YOUR AGENT
Google Gemini
Google Gemini is Google’s flagship multimodal model family and assistant, delivered as a hosted service rather than as downloadable weights.
At a glance
Google Gemini is Google’s flagship multimodal model family and assistant, delivered as a hosted service rather than as downloadable weights.
Catalog categories: ai_model.
Catalogued capability areas: - Multimodal reasoning (text, image, audio, video) - Long-context generation and tool use - Hosted assistant plus a developer API
This summary reflects the catalog record, not a complete product specification. Verify intended users, end-to-end workflows, supported platforms, edge cases, data handling, and constraints against cited sources. Treat capability labels as research leads until that verification is complete.
- Multimodal reasoning (text, image, audio, video)
- Long-context generation and tool use
- Hosted assistant plus a developer API
Open-source alternative
2 candidate projects in the landscape. A candidate may cover all, part, or none of what Google Gemini does — read the coverage and gaps below.
Llama / Mistral / Qwen / Gemma
mixed (mostly open weights)Capable open-weight model families you can run yourself.
Many are open-weight under custom licences, not OSI open source.
Ollama
MITLocal model runner and OpenAI-compatible API.
Runs the weights; does not provide a frontier model.
Coverage and gaps
Alternative coverage
Partial. Self-hostable model runners and open-weight families exist and are improving fast, but a fully open, frontier-equivalent model is pending.
The gap
No open project matches a frontier hosted model on capability; open-weight families trail the hosted leaders and often carry use restrictions.
Build this product with your agent
This Markdown file is the coding-agent skill for Google Gemini. Give it to your agent in your project; it guides research, reuse decisions, implementation and verification.
- agent-kit.mdDownloadable Markdown skill with product-specific instructions for your coding agent.available
- skill.mdGenerated copy of the same product skill — rae google-gemini --out ./google-gemini-agent.in kit
- mcp.jsonOptional MCP server config — rae google-gemini --out ./google-gemini-agent.in kit
- Product workspaceOptional starting workspace for your agent — rae build product google-gemini --out ./google-gemini.
- ChoosePick the product you want to build.
- ReviewAsk your agent to check the product skill, evidence and alternatives.
- BuildGive the skill to your coding agent and build in your own project.
- VerifyRun the checks for your implementation and review the result.
Evidence
Sources
- https://huggingface.co/blog/daya-shankar/open-source-llm-models-to-run-locally
- https://www.instaclustr.com/education/open-source-ai/top-7-open-source-llms-for-2026/
- https://github.com/12britz/awesome-free-models
Provenance
- Catalog record — catalog/products.jsonl
- Research dossier — research/companies/google/products/google-gemini.md
- REA web-surface report — catalog/rea/google/google-gemini.md
- Open-source skill file — catalog/skills/google/google-gemini.md
Next step
Give the Google Gemini skill to your coding agent in your project. Ask it to evaluate Llama / Mistral / Qwen / Gemma ↗ against your needs, reuse what fits, then follow the skill's build and verification instructions.
Give the product skill to your coding agent in the project where you want to build. Optional starter workspace: rae build product google-gemini --out ./google-gemini. CLI reference.