Meet Gemma 4 12B! A unified, encoder-free multimodal model designed to bring high-performance intelligence directly to your laptop, and released under an Apache 2.0 license. Bridging the gap between edge efficiency and advanced reasoning. Here is what’s new with Gemma 4 12B: 👇

Bittensor Ecosystem Highlights :: June 15–21, 2026 SUBNET ACHIEVEMENTS [ @vidaio_ - SN85 ] Vidaio ranked #1# for compression efficiency in its benchmark against major cloud encoders, with lower pricing and one-pass target matching. > [ @webuildscore - SN44 ] @MaxScore was on stage at @VivaTech with @jtledore from PwC France and Cémoi, sharing how Score is deploying its decentralized Vision AI in one of Cémoi’s main factories in France. > @manakoai unveiled the video presentation of their upcoming vision AI product. > Score also released their new vision skill “Detect-road-signs”. > [ @chutes_ai - SN64 ] Chutes finished moving the entire platform onto TEE and brought its first Blackwell B200s online. > Chutes is now also a provider on @openGPUnetwork. > [ @numinous_ai - SN6 ] Numinous is integrating its predictive market signals into @arespro to surface high-impact news and market context for traders. > [ @theminos_ai - SN107 ] Minos can now generate a whole synthetic genome in under 20 minutes on eight H200 GPUs from @lium_io, roughly 440x faster than CPU workflows. > Minos was accepted into the Cloudflare for Startups program to support the next phase of its decentralized genomic infrastructure. > [ @EnigmaSN63 - SN63 ] Enigma’s second RSA milestone was solved, factoring 460-bit RSA in 3.9 hours. > [ @QuantumSN48 - SN48 ] @PennyLaneAI is now integrated into Open Quantum, making it easier to run quantum ML and optimization workflows on real quantum hardware. > [ @yanez__ai - SN54 x @bitmind - SN34 ] Yanez and BitMind stress-tested their AI detection models on 300K adversarial face images, reaching 98% overall accuracy. > [ @QuasarModels - SN24 ] Quasar-Preview trended on Hugging Face, reaching page two alongside major open-source AI models. > [ @EndureNet - SN30 ] They announced @ForgeLending, the first product built on Endure, a native money market for Bittensor. > [ @TrajectoryRL - SN11 ] TrajectoryRL showed that frontier-level models can be served at scale on consumer GPUs, using Qwen3-6.35B-A3B with 256K context. > [ @heydittoai - SN118 ] DittoBench is live, benchmarking agents on memory, tool use and speed. > [ @blockmachine_io - SN19 ] Blockmachine’s June buyback nearly doubled May’s, reaching $3,981. > [ @ai_detection - SN32 ] It’s AI added OCR support for scans, photos, PDFs and broken encodings. > SUBNET LAUNCH [ @RalphLabsAI - SN40 ] Ralph launched as a decentralized research network where autonomous agents compete to improve a shared, proof-tested AI training recipe. > [ @CookingTao_ - SN122 ] @LamidaGlobal announced CookingTAO, a subnet aiming to make it easier to deploy, manage and monetize miners across Bittensor. > BITTENSOR ECOSYSTEM [ @iclblockchain ] @imperialcollege is co-hosting the UK AI Agent Hackathon this summer, with Opentensor, Macrocosmos and other ecosystem partners involved. > [ @taoswap_org ] Taoswap launched its redesign, with a refreshed UI, new comparison tools and more subnet trading metrics. > [ @taodotcom ] They highlighted their bridge app, letting users wrap native TAO and bridge it to Solana and other chains. > [ @nametensor ] NameTensor launched a TNS Migration Program with discounts for TAO Name Service holders. > PODCASTS @opentensor Novelty Search hosted by @const_reborn. > @IOV_OWL podcast with @MaxScore from Score. > @markjeffrey Hash Rate podcast with @peytonspencer from Ditto. > @markjeffrey Hash Rate podcast with @gavinzaentz from @LeadpoetAI. > @gordonfrayne podcast with @centrum_blue from @theminos_ai. > @gordonfrayne podcast with @sebyrubino from Zipcode. > @gordonfrayne podcast with @yubrew from Bitsec. > @jollygreenmoney podcast with @Tom_dot_b from Bitcast. >

Attention is a lookup. Each token builds a query, compares it against every key in the sequence, and pulls value vectors weighted by the match. Stack that 96 layers deep and you get a frontier model. Video covers the full pipeline: Q/K/V, attention scores, encoder blocks.

Supercharge your Z-Image-Turbo workflows with the new Z-Image-Engineer-V6! Built specifically for Z-Image-Turbo, it turns simple prompts into cinematic descriptions and directly upgrades image depth, texture, and quality as a swappable text encoder.

Here’s this week’s shipping recap 👇 — Nano Banana 2 & Nano Banana Pro are now GA and available via the Gemini Enterprise Agent Platform, Gemini API, and in @GoogleAIStudio —Co-Scientist, our new multi-agent system for structured scientific thinking, generates and refines novel hypotheses to solve complex scientific problems — dreambeans from @GoogleLabs works overnight to curate a personalized daily collection of topics that are relevant to you based on your connected Google apps — @GoogleGemma 4 12B, our unified encoder-free model, brings powerful multimodal intelligence straight to your laptop fully offline — Gemma 4 models and their drafters are now optimized with Quantization-Aware Training (QAT) to cut memory requirements and maximize on-device performance — @GoogleMagenta RealTime 2 is our open-weights live music model that you can play like an instrument using a MIDI keyboard, text prompts, and gestures

Today we’re introducing Gemma 4 12B — our latest open model that brings advanced agentic reasoning, vision and audio directly to your laptop. It delivers performance nearing our larger Gemma models with a much smaller total memory footprint, while being small enough to run locally with just 16GB of VRAM. It’s open and accessible for everyone to use under a permissive Apache 2.0 license. This is all made possible by our new, unified architecture that removes separate multimodal encoders. Here’s how we did it 🧵

Kimi K2.5 tech report just dropped! Quick hits: - Joint text–vision training: pretrained with 15T vision-text tokens, zero-vision SFT (text-only) to activate visual reasoning - Agent Swarm + PARL: dynamically orchestrated parallel sub-agents, up to 4.5× lower latency, 78.4% on BrowseComp - MoonViT-3D: a unified image–video encoder with 4× temporal compression, enabling 4× longer videos in the same context - Toggle: token-efficient RL, 25–30% fewer tokens with no accuracy drop Here's our work toward scalable, real-world agentic intelligence. More details in the report 👉