All AI models
Open model familyVerified October 23, 2026

Google

Gemma 4

Gemma 4 is a flexible family to explore when you want a Google-backed open-model route with options across model sizes and specialized variants. It is more useful as a selection set than as one universal answer: first choose the right checkpoint, then evaluate its license, modality, and hardware fit.

What changed: Checked October 23, 2026: Google's release page lists no Gemma 5. The newest Gemma release is EmbeddingGemma 2 (740M parameters, October 6, 2026), an embedding model rather than a chat model.
License
Apache 2.0 (per model card)
Context
128K (small models) to 256K (medium models)
Modalities
Text · Unified and specialist variants
Parameters
E2B, E4B, 31B, and 26B A4B (March 31, 2026), plus 12B Unified (June 3, 2026)

What the license means

Google's Gemma 4 model card on Hugging Face lists the license as apache-2.0 and links Google's Gemma 4 license terms. Read both, plus the use policy, before offering a hosted service or redistributing a modified checkpoint.

Open-weight, open source, and no-cost API access mean different things. Always review the linked terms and your own data, hosting, and compliance requirements before shipping.

API and operating cost

Input / 1M
No universal Gemma API rate
Output / 1M
No universal Gemma API rate

Open model weights do not create a single provider API price. Compare the cost of your selected runtime or hosting provider instead.

Check current provider price

Where it fits

  • Right-sized open-model experiments
  • Specialist model selection
  • Google AI ecosystem prototyping

Where it does not fit

  • Assuming every Gemma 4 variant has the same capabilities or context window
  • Budgeting from a generic API price

Local deployment reality

Gemma spans small and larger variants, so the practical local path depends on the checkpoint and quantization. Treat it as a family selection problem rather than a one-model hardware estimate.

Runtime depends on selected checkpoint

Practical setup notes

  • Choose the specific Gemma 4 checkpoint before estimating memory or inference cost.
  • Read the Gemma 4 license for your distribution model, especially if you offer hosted access.

Capability notes

Reasoning
Model variant dependent
Tools and output
Google AI and open ecosystem tooling vary by checkpoint

Limitations to plan around

  • The family spans multiple sizes and use cases, so broad claims are not useful.
  • There is no one provider-hosted API rate for all Gemma models.

Continue comparing

Related model decisions

See all reviewed models
Open source

Qwen3.6

Local coding assistance

Open-weight

gpt-oss-20b

Local text reasoning

Open source

Mistral Small 4

Multimodal document analysis

Editorial note: This is a dated decision page, not a benchmark leaderboard or a substitute for legal, security, or infrastructure review. Model and pricing facts can change after October 23, 2026; use the linked primary sources before committing budget or customer data.Rechecked October 23, 2026: family members, release dates, context windows, and the license listed on the model card. Tooling notes date from the August 11, 2026 audit.