- 🧩 MoE with 25.2B total / 3.8B active parameters
- 🔒 Runs in a Trusted Execution Environment with hardware attestation
- 👁️ Multimodal input across text and images
- 📏 64K-token context window
- 🌍 Uncensored variant with reduced content refusals
- ⚡ fp8 quantization for efficient serving
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 4 26B A4B Uncensored is a privacy-hardened, uncensored build of Google's Gemma 4 mixture-of-experts model, running inside a Trusted Execution Environment (TEE) that produces hardware attestation evidence for independent verification. Architecturally it pairs 25.2B total parameters with only 3.8B active per token, keeping inference lightweight while retaining broad knowledge, and it accepts both text and image input across a 64K-token context, served at fp8 precision.
Released in May 2026, it sits alongside the standard Gemma 4 Uncensored as the confidential-compute counterpart, and joins Google's TEE-backed lineup that also includes Gemma 4 31B Instruct and the earlier Gemma 3 27B. The "uncensored" tuning loosens the default safety guardrails, while the sparse MoE design gives it a favorable speed-to-capability ratio compared with dense models of similar footprint.
This build is best suited to users who want open-ended, low-refusal generation with verifiable end-to-end privacy — creative writing, roleplay, and sensitive research where both confidentiality and multimodal understanding matter, without paying the latency cost of a fully dense model.
This About section is AI-generated from public sources via VeniceStats + Venice inference, with no human editing. It may contain inaccuracies.
Daily usage · Gemma 4 26B A4B Uncensored · Organic sellers
| Seller | Health | Rep↓ | Sales | Tokens | Routing | Input $/M | Cached $/M | Output $/M | Save |
|---|---|---|---|---|---|---|---|---|---|
| Phala 0x88c8…15d6 | healthy | 89.61 | $6.45 | 37.91M | #1 | $0.15 | $0.05 | $0.70 | −21% |
| RedPill 0x7e99…dc4f | healthy | 43.30 | $0.00 | 2.00k | gated | $0.15 | $0.05 | $0.70 | −21% |
"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.