Google Cloud Vertex AI tuning (Agent Platform)
Google LLC; Google now calls the product Generative AI on Gemini Enterprise Agent Platform (formerly Generative AI on Vertex AI) · cloud.google.com · Last checked:
- Managed fine-tuning API
- Cloud platform fine-tuning
- Reinforcement fine-tuning
Company status
Active. Business line of Google Cloud; parent Alphabet Inc. is publicly listed.
Sources 1, 2, 3, 4, 5 · checked 2026-11-08
Capabilities
- Supervised fine-tuning for Gemini models (3.5 Flash-Lite, 3.5 Flash, 3.1 Flash-Lite, 2.5 Pro, 2.5 Flash-Lite, 2.5 Flash)
- Preference tuning for Gemini 2.5 Flash-Lite and 2.5 Flash
- Reinforcement learning fine-tuning (LoRA adapter against reward functions) for Gemini 3.5 Flash and 3.1 Flash-Lite
- Supervised fine-tuning of open models (Gemma 3, Llama 3.x and 4, Qwen 3) with per-token pricing
- Parameter-efficient (adapter) tuning and full fine-tuning described
Sources 1, 2, 3, 4, 5 · checked 2026-11-08
Who it tends to fit
- Teams standardizing on Gemini that want tuned Gemini models
- Google Cloud customers needing HIPAA coverage under a BAA
Sources 1, 2, 3, 4, 5 · checked 2026-11-08
Pricing
Supervised fine-tuning per 1,000 training tokens: Gemini 2.5 Pro $0.025; Gemini 3.5 Flash $0.01 (SFT and RL fine-tuning); Gemini 2.5 Flash $0.005; Gemini 3.5 Flash Lite and 3.1 Flash Lite $0.003; Gemini 2.5 Flash Lite $0.0015. Open models per 1M training tokens, for example Llama 3.1 8B $0.67, Gemma 3 27B IT $6.83, Qwen 3 32B $6.57. Training tokens = dataset tokens x epochs.
Sources 1, 2, 3, 4, 5 · checked 2026-11-08
Published certifications
- SOC 2 (Generative AI on Agent Platform in scope)
- ISO/IEC 27001 (in scope)
- HIPAA (Generative AI on Agent Platform listed; BAA required)
Listed as stated by the vendor. Scope can differ by product or subsidiary; request current reports.
Sources 1, 2, 3, 4, 5 · checked 2026-11-08
Limitations to check
- Tuned Gemini models are served in Google Cloud; Gemini weights are not exportable
- From Gemini 3, tuned model endpoint prediction price is 1.5x the base model in non-global regions (per pricing page)
- Method support differs by model; check the tuning table before planning
Sources 1, 2, 3, 4, 5 · checked 2026-11-08
Sources
- Supported tuning methods (SFT, preference tuning, RLFT) and the Gemini models for each; adapter vs full fine-tuningdocs.cloud.google.com/vertex-ai/generative-ai/docs/models/tune-modelsChecked 2026-11-08
- Model tuning price table for Gemini and open models; tuned endpoint 1.5x notecloud.google.com/vertex-ai/generative-ai/pricingChecked 2026-11-08
- Generative AI on Gemini Enterprise Agent Platform (formerly Generative AI on Vertex AI) in SOC 2 scopecloud.google.com/security/compliance/soc-2Checked 2026-11-08
- Generative AI on Gemini Enterprise Agent Platform in ISO/IEC 27001 scopecloud.google.com/security/compliance/iso-27001Checked 2026-11-08
- Generative AI on Agent Platform listed in HIPAA covered productscloud.google.com/security/compliance/hipaaChecked 2026-11-08
Report a correction: [email protected]. Facts are re-checked at least every 90 days.