- 🔒 Runs in a Trusted Execution Environment with hardware attestation.
- 🧠 Google's 27B open-weight multimodal vision-language model.
- 👁️ Adds native image understanding via a SigLIP vision encoder.
- 🌐 Understands 140+ languages using a Gemini-derived tokenizer.
- 📏 Natively supports 128k context; Venice exposes a 40,000-token window.
- 🆕 Big jump from Gemma 2: multimodality, longer context, RoPE rescaling.
- 🔧 Adds function calling and structured outputs.
- 🔒 Attestation evidence available for independent verification.
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 →Gemma 3 27B is the largest model in Google's Gemma 3 open-weight family, a text-and-image-input, text-output vision-language model with roughly 27 billion parameters. In this Venice deployment it runs inside a Trusted Execution Environment (TEE) with hardware attestation, so users can independently verify that prompts are processed in an encrypted, isolated enclave — the privacy layer, rather than a change to the underlying weights. Venice exposes it with a 40,000-token context window, though the base model natively supports up to 128k tokens.
Relative to its own predecessor, Gemma 2, the generational gains are substantial. Google added multimodality: the 4B, 12B and 27B models employ a custom SigLIP vision encoder that lets them interpret images and short video. Context length grew from Gemma 2's 8k to 128k via RoPE rescaling (global-layer base frequency raised from 10k to 1M), and a new Gemini-style tokenizer plus revised data mixture improved coverage across 140+ languages. Google reports enhanced math, coding and instruction-following, with architectural changes cutting KV-cache memory during long-context inference.
On Hugging Face's reported benchmarks, the 27B instruction-tuned model scores 67.5 on MMLU-Pro and around 69 on MATH, alongside multimodal scores like 64.9 on MMMU. Within Google's broader lineup this model has since been succeeded by newer generations, including Gemma 4 31B Instruct and Gemma 4 26B A4B Uncensored, which offer further architectural updates. Gemma 3 27B remains a capable, resource-efficient open option for chat, document analysis and vision-language tasks.
This About section is AI-generated from public sources via VeniceStats + Venice inference, with no human editing. It may contain inaccuracies.
Daily usage · Gemma 3 27B · Organic sellers
| Seller | Health | Rep↓ | Sales | Tokens | Routing | Input $/M | Cached $/M | Output $/M | Save |
|---|---|---|---|---|---|---|---|---|---|
| ▲ Apex Ant 0x73b4…e736 | healthy | 81.71 | $0.01 | 110.84k | #1 | $0.06 | $0.06 | $0.10 | −57% |
| OpenCheap 0x0ee1…6c41 | — | 12.43 | $0.00 | 59.94k | gated | $0.0588 | $0.0588 | $0.098 | −58% |
| Apex TEE Test 0xe672…7955 | — | 0.00 | — | — | gated | $0.096 | $0.063 | $0.54 | −31% |
"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.