AI News: 6 New Models & A Nano Banana Killer

K3Krea 3KreaAvailable in EU

Image generation and editing model in public beta. The creator demoed inline edits (recolor eyes, change text and logos) in about 10 seconds, and found it cleaner than ChatGPT Image or Nano Banana. He named it the image winner of the week.

F3FLUX 3 Image KleinBlack Forest LabsAvailable in EU

Open-weight image model with weights on Hugging Face. It runs on consumer GPUs like an RTX 4070 and on a DGX Spark, and can be tried free on Spaces. The creator found it decent but not special apart from being open.

QIQwen Image 2.1 TurboAlibabaAvailable in EU

Open-weight accelerated checkpoint of Qwen Image 2.1, trading some quality for speed. It can be tested on Hugging Face and also does image editing.

GDGenspark DesignGensparkUnverified

Sponsored AI design tool. It turned Future Tools screenshots into a full mobile app design with layered, editable screens, exportable components and design tokens.

G4Gemini 4 CarbonGoogleUnverified

Google's new model, roughly caught up with GPT-6 Astra and Claude Opus 5.5 on most benchmarks. API price is $2/$10 per million tokens until end of November. The creator tested it on Beauc Bench, Mega Bonk and a Gary Busey 3D-printable STL bust.

MWMuse WonderMetaUnverified

Meta's new state-of-the-art model at $4/$20 per million tokens, scoring 79% on a coding benchmark and near top-tier on Artificial Analysis. The creator's tests showed a working Mega Bonk game with weaker graphics, and ranked 14th on his Beauc leaderboard.

MSMuse Spark OpenMetaUnverified

Open-weight release of Muse Spark, 816B parameters with 1M context. Meta calls it the most capable US open-weight model, and it was optimized with Apple to run on an M5 Ultra Mac Studio. It tops most open-weight comparisons except Kimi on the coding benchmark.

MBMAI Base 2MicrosoftUnverified

Microsoft's newly pre-trained model with 80B active parameters, claimed to outperform every open-weight model tested. Post-training isn't finished and it isn't available, so the creator couldn't test it.

NNNVIDIA Nemotron 3.5 Super VLNVIDIAUnverified

120B open-weight model scoring 26 on Artificial Analysis, below DeepSeek, Qwen and GLM. Its selling point is high output speed.

U1Ultramarine 1 FlashPaper InstrumentsAvailable in EU

Open-weight model claimed to match Fable class on knowledge work at 60x lower cost. Pricing is $0.15 input and $0.50 output per million tokens. The creator couldn't benchmark it because it wasn't on OpenRouter yet.

OCOpenAI Codex updates (28-day shipping challenge)OpenAIAvailable in EU

Weekly recap of OpenAI's 28-day Codex/Work shipping event. Updates included composer predictions, creating a dot from the phone, delegating Codex work to a dot, a 500K default context for GPT 6.1 Soul, a 60% disk storage reduction and more compute. The creator called it a minor week.

MDMicrosoft Decision 1MicrosoftUnverified

Decision model like Jev that returns calibrated probability scores for yes/no, multiple-choice and rating options, plus rubric grading of AI responses and agent actions. Claimed slightly higher accuracy.

NINimble-v4Bespoke LabsUnverified

Decision models, Nimble-v4 and v4-max, that support multiple choice, yes/no and rubric scores. Claimed to outperform Jev on shared benchmarks.

GVGrok Voice 3.0xAIUnverified

New voice model with a free sandbox. The creator built a Future Tools support voice agent that answered from site content with low latency, and showed some interruption-handling quirks.

EIElevenLabs Instant SoundtracksElevenLabsAvailable in EU

Feature in Eleven Creatives that watches an uploaded video and generates a soundtrack with synchronized sound effects.

FAFacebook AI feed controlsMetaUnverified

Facebook update that lets users enter their interests with AI to tune the feed, with ranking adapting in real time. The creator didn't test it.

CTCircle to Search Scam DetectionGoogleUnverified

Android feature where you long-press a text message to trigger Circle to Search, and an AI overview flags scam red flags. The creator couldn't test it.

Agent orchestration tool where multiple agents and people work together in a shared chat-style messaging system. The creator saw a demo and found it impressive.

Experimental Xbox division exploring AI and reverse-engineering for game mashups, decompiling and AI-generated games.

ATApple TV 4KAppleLimited in EU

New Apple TV 4K unveiled at Apple's event, with Siri built in.

AHApple HomeViewAppleNot in EU

Apple's countertop smart display with Siri, aimed at AI use with the smart home. The creator compared it to a bigger Apple Watch or an Alexa screen and hasn't used it.