Google Gemma 3 27B Instruct
gemma-3-27b-itgoogle-gemma-3-27b-instructgoogle-gemma-3-27b-it- 🧠 27B open-weight instruct model, successor to Gemma 2
- 📏 Handles context windows up to 128k tokens
- 👁️ Vision-language input with text outputs
- 🔧 Function calling, structured outputs, and web search
- 🌍 Understands over 140 languages
- 📜 Released under Google's Gemma license, fp8 quantized
Google is an American multinational technology corporation and one of the world's most valuable brands. A subsidiary of parent company Alphabet Inc., Google operates across search, cloud computing, consumer electronics, and artificial intelligence. Its DeepMind and Google…
Explore 28 more models by Google →Google Gemma 3 27B Instruct is Google's 27-billion-parameter open-weight model and the release that brought multimodality to the Gemma line, adding vision-language input on top of the family's text generation. It handles context windows up to 128k tokens, understands more than 140 languages, and improves on math, reasoning, and chat over its Gemma 2 predecessor, while supporting structured outputs and function calling. Served here in fp8, it pairs a manageable footprint with capable general-purpose performance, reflected in its heavy adoption on Hugging Face.
Within Google's lineup this is the prior generation of the flagship Gemma instruct line: it has since been joined by Gemma 4 31B Instruct (released April 2026) and the mixture-of-experts Gemma 4 26B A4B Instruct. It sits alongside Google's broader catalogue of Gemini text models, Veo video, Lyria music, and Nano Banana image generators, but remains the accessible open-weight option many users still reach for.
Best suited for developers who want a dependable, openly licensed multimodal model for multilingual chat, vision tasks, tool-calling agents, and structured-output workflows without committing to a larger flagship.
This About section is AI-generated from public sources via VeniceStats + Venice inference, with no human editing. It may contain inaccuracies.
Daily usage · Google Gemma 3 27B Instruct · Organic sellers
Top sellers · Google Gemma 3 27B Instruct · Organic
- Venice.ai Proxy91.9%
- JOI SEED5.6%
- Fire Ant 🔥🐜2.4%
- Swift Ant0.2%
| Seller | Health | Rep↓ | Sales | Tokens | Routing | Input $/M | Cached $/M | Output $/M | Save |
|---|---|---|---|---|---|---|---|---|---|
| Venice.ai Proxy 0x1f22…18c9 | healthy | 86.84 | $0.10 | 1.40M | gated | $0.066 | $0.066not advertised | $0.11 | −17% |
| JOI SEED 0x808a…6a8b | — | 0.07 | $0.01 | 154.46k | gated | $0.0488 | $0.0488 | $0.122 | −39% |
| surplus-hive 0x6145…5c58 | — | 0.00 | — | — | gated | $0.2917 | $0.2917 | $0.4861 | — |
| Swift Ant 0xe49f…a3a9 | — | 0.00 | $0.00 | 1.89k | gated | $0.1058 | $0.1058 | $0.1764 | — |
| Fire Ant 🔥🐜 0xbe05…bc5d | — | 0.00 | $0.00 | 25.01k | gated | $0.0996 | $0.0996not advertised | $0.1664 | — |
| surplus-hive 0xe177…e91c | — | 0.00 | — | — | gated | $0.0908 | $0.0908 | $0.1519 | — |
| AntFeed 0xddb6…1442 | — | 0.00 | — | — | gated | $0.088 | $0.088not advertised | $0.176 | — |
| antseed-neon-puma-944e 0x6650…944e | — | 0.00 | — | — | gated | $0.0409 | $0.0409not advertised | $0.0682 | −49% |
| Joi Gateway 0x8a90…7007 | — | 0.00 | — | — | gated | $0.039 | $0.039 | $0.0976 | −51% |
| Leftermute 0x388b…5389 | — | 0.00 | — | — | gated | $0.0127 | $0.0127not advertised | $0.0455 | −84% |
"Best price" and the seller table are live Antseed catalog data (advertised $/1M tokens — or $ per generated image for unit-billed image models — not settled amounts). Reputation = buyer trust score (0-100, the Antseed SDK's own formula). "Routing" = the SDK's default buyer routing (what the VPR desktop app ships with): a trust ≥ 60 gate on the effective reputation, then cheapest-first among routable sellers; live failover state (per-peer cooldowns) is buyer-side runtime and not included. Model knowledge (TLDR, provider, About) via the VeniceStats enrichment layer. Advertised catalog, not the model used in any specific purchase. "Usage on Antseed" counts only settlements whose buyers share the per-model split on-chain (metadata v2/v3, opt-in), so every usage figure is a lower bound.