The Department of Homeland Security has ordered all federal civilian agencies to stop using DeepSeek products and remove any existing deployments from their IT systems. The order also bars the Chinese AI lab from taking part in federal government contracts. DHS warned that DeepSeek could transfer sensitive data on US government employees and vendors to the Chinese government and that its models could contain malware or security vulnerabilities. The move came days after DeepSeek released its V4.1 Flash model.
Anthropic plans to allocate 5% of its shares to a new Public Benefit Fund, subject to stockholder approval. The board could add another 2% for each 50% valuation increase above a baseline, with worker disruption among the potential uses. Including earlier founder and employee pledges, the company says roughly 20% of its shares would support public benefit efforts.
AnthropicQuick look +
Anthropic plans to allocate 5% of its shares to a new Public Benefit Fund, subject to stockholder approval. The board could add another 2% for each 50% valuation increase above a baseline, with worker disruption among the potential uses. Including earlier founder and employee pledges, the company says roughly 20% of its shares would support public benefit efforts.
A Super Mario 64 demake running on Game Boy Advance hardware: a 16.8 MHz ARM7 with no floating-point unit, four playable levels, software 3D and Mario animated with the game's own animation data. The creator picked the project and the trade-offs; Claude wrote the code, the tests and the tooling.
GamesQuick look +
A Super Mario 64 demake running on Game Boy Advance hardware: a 16.8 MHz ARM7 with no floating-point unit, four playable levels, software 3D and Mario animated with the game's own animation data. The creator picked the project and the trade-offs; Claude wrote the code, the tests and the tooling.
The repo ships nothing from Nintendo. An optional import reads your own ROM to build Mario's real model and level meshes, so the project stays free and MIT-licensed.
MiniMax M3.1 Flash got more questions right than its rivals and still scored a 93% hallucination rate. What the number means, why accuracy does not cancel it out, and how to test a cheap model on your own work before you move to it.
Our analysis · written by handQuick look +
What does a model’s hallucination rate actually measure?
MiniMax M3.1 Flash got more questions right than its rivals and still scored a 93% hallucination rate. What the number means, why accuracy does not cancel it out, and how to test a cheap model on your own work before you move to it.
Eight states have filed separate lawsuits against OpenAI, alleging that ChatGPT harms young users and that the company misled families about its safety. Iowa announced its case alongside Kansas, Nebraska and West Virginia, with four other states also suing. The cases seek changes through state consumer protection laws; the accusations are allegations, rather than findings by a court.
Iowa Attorney GeneralQuick look +
Eight states have filed separate lawsuits against OpenAI, alleging that ChatGPT harms young users and that the company misled families about its safety. Iowa announced its case alongside Kansas, Nebraska and West Virginia, with four other states also suing. The cases seek changes through state consumer protection laws; the accusations are allegations, rather than findings by a court.
A creator used Claude through Blender MCP to model a Matrix-inspired apartment from scratch, without image-to-3D tools. The project took roughly five days part-time, with extensive annotations and art direction from the creator.
3D, VR & spatialQuick look +
A creator used Claude through Blender MCP to model a Matrix-inspired apartment from scratch, without image-to-3D tools. The project took roughly five days part-time, with extensive annotations and art direction from the creator.
Claude handled modeling, UVs, cameras, materials and compositing. The creator stresses that solid 3D fundamentals and iterative feedback mattered more than a single prompt.
The OpenCode Console exposure, read as a checklist: revoke and reissue keys from the provider, check usage, cap spend, and decide where a coding agent should keep credentials in the first place.
Our analysis · written by handQuick look +
What should I do if a coding agent that stored my API keys is breached?
The OpenCode Console exposure, read as a checklist: revoke and reissue keys from the provider, check usage, cap spend, and decide where a coding agent should keep credentials in the first place.
Liquid AI released LFM2.5-24B-A2B, a reasoning model with 24 billion total parameters, 2 billion active per token and a 131,072-token context window. The company reports 184 tokens per second on an Apple M5 Max and stronger results than larger rivals on several tool-use tests, while acknowledging weaker coding and knowledge scores. Downloadable weights and support for MLX and llama.cpp make it a practical option for private agents running locally.
Liquid AIQuick look +
Liquid AI released LFM2.5-24B-A2B, a reasoning model with 24 billion total parameters, 2 billion active per token and a 131,072-token context window. The company reports 184 tokens per second on an Apple M5 Max and stronger results than larger rivals on several tool-use tests, while acknowledging weaker coding and knowledge scores. Downloadable weights and support for MLX and llama.cpp make it a practical option for private agents running locally.
A former personal trainer who writes zero code built a 3D Vampire Survivors style spell game in 2 days with Claude Code, Higgsfield 3D and ElevenLabs. Five minute runs, aimed spells, and the Lich King shows up at the 4 minute mark.
GamesQuick look +
A former personal trainer who writes zero code built a 3D Vampire Survivors style spell game in 2 days with Claude Code, Higgsfield 3D and ElevenLabs. Five minute runs, aimed spells, and the Lich King shows up at the 4 minute mark.
He made it to play on a second monitor while waiting in Deadlock queues. His first attempt, an idle game, was bad, so he scrapped it and started over.
Check the original project, the input-to-output workflow, the role of AI and the limits of the demonstration. A practical guide to separating working prototypes from memes, hook templates and promotional clips.
Our analysis · written by handQuick look +
How can I tell whether an AI demo shows a real working project?
Check the original project, the input-to-output workflow, the role of AI and the limits of the demonstration. A practical guide to separating working prototypes from memes, hook templates and promotional clips.
Google rolled out Gemini 4 Argon globally through the Gemini API, to Google AI Pro subscribers in the Gemini app and AI Mode, and in its Antigravity coding tool for Ultra subscribers. The model is priced at an introductory $2 per million input tokens and $10 per million output tokens until November 30, after which the rate doubles to $4 and $20. Argon had been restricted since its late September announcement, when Google limited early access to trusted cybersecurity users over safety concerns. There is no free tier, so only paying users and developers can use it.
VentureBeatQuick look +
Google rolled out Gemini 4 Argon globally through the Gemini API, to Google AI Pro subscribers in the Gemini app and AI Mode, and in its Antigravity coding tool for Ultra subscribers. The model is priced at an introductory $2 per million input tokens and $10 per million output tokens until November 30, after which the rate doubles to $4 and $20. Argon had been restricted since its late September announcement, when Google limited early access to trusted cybersecurity users over safety concerns. There is no free tier, so only paying users and developers can use it.
Hive Robotics is training a Unitree G1 humanoid to restock shelves inside a real Føtex supermarket in Lyngby, Denmark, with shoppers walking past it. The G1 lists at around 16,000 dollars.
Robotics & hardwareQuick look +
Hive Robotics is training a Unitree G1 humanoid to restock shelves inside a real Føtex supermarket in Lyngby, Denmark, with shoppers walking past it. The G1 lists at around 16,000 dollars.
Right now a human still guides the training. The goal is for it to restock autonomously by early 2027.
A practical test plan for agents that read documents, browse websites or use tools. Check unauthorized actions, useful task completion and permission boundaries with harmless fixtures.
Our analysis · written by handQuick look +
How can I test whether my AI agent follows malicious instructions in retrieved content?
A practical test plan for agents that read documents, browse websites or use tools. Check unauthorized actions, useful task completion and permission boundaries with harmless fixtures.
In October 2026, Linus Torvalds published an experimental eight-patch Linux branch whose code, tests and first draft description were written with Claude Code under his direction. On his laptop it tripled one kind of read. The same weekend, AI tools helped him find three bugs that had sat in the kernel since 2013. Here is what was built, what it does and does not prove, and the rules the kernel sets for AI-assisted code.
Our analysis · written by handQuick look +
What did Linus Torvalds use Claude Code for, how much faster is the result, is it in Linux yet, and what are the kernel's rules for code written with AI?
In October 2026, Linus Torvalds published an experimental eight-patch Linux branch whose code, tests and first draft description were written with Claude Code under his direction. On his laptop it tripled one kind of read. The same weekend, AI tools helped him find three bugs that had sat in the kernel since 2013. Here is what was built, what it does and does not prove, and the rules the kernel sets for AI-assisted code.
On October 12, 2026, Meta released Muse Wonder and Muse Wonder Full, its first models built to compete at the frontier. Wonder is on the Meta Model API at $4 and $20 per million tokens. Wonder Full, the stronger one, ties Claude Sonnet 5.5 on the independent Intelligence Index but is only in Muse Code. Here is what each scores, what it costs, and when to pick it.
Our analysis · written by handQuick look +
What are Meta's Muse Wonder and Muse Wonder Full, how do they compare with Claude, GPT-6 and Gemini, what do they cost, and which one can you actually use?
On October 12, 2026, Meta released Muse Wonder and Muse Wonder Full, its first models built to compete at the frontier. Wonder is on the Meta Model API at $4 and $20 per million tokens. Wonder Full, the stronger one, ties Claude Sonnet 5.5 on the independent Intelligence Index but is only in Muse Code. Here is what each scores, what it costs, and when to pick it.
Tools can appear in more than one category, so these add up to 632, not 404.
Picked by Gaurav at The Singularity Labs, a creative studio that also makes the shorts on @aitickerdaily(105 subscribers, 293 videos, as of Oct 11). Studio work never earns a place here.
ElevenLabs' most expressive voice model, with a Turbo version fast enough for live agents
NewAdded Oct 19, 2026
Eleven v4 and Eleven v4 Turbo are ElevenLabs' text-to-speech models launched September 28, 2026. v4 is the expressive model for narration and long-form audio; Turbo is the real-time version, with a median inference latency of about 100 ms and about 150 ms to first speech, per ElevenLabs. Both cover 90+ languages, take inline audio tags such as [whispers] or [door slams] to steer delivery and add sound effects, and keep pacing steady across long scripts. Instant Voice Clones work from ten seconds of audio, and Professional Voice Clones, unavailable on v3, are supported again. Generations run up to 10,000 characters. Both are available in the ElevenLabs apps, through the API by switching model_id, and in ElevenAgents.
Why: It replaces v3 as the model to reach for in ElevenLabs, and the split matters: one model for performance, one for conversation, instead of choosing between expressive and fast. ElevenLabs reports it at the top of Artificial Analysis's speech leaderboard; that is the company's citation, and the leaderboard snapshot will confirm or correct it.
Suno's new generation of music models, built with Warner Music Group, BMG and Believe
NewAdded Oct 19, 2026
Suno v6, announced September 9, 2026, is three music models: v6, the flagship for precise, polished songs; v6-wild, for less predictable and more varied output; and v6-mini, a faster, more efficient model available on every plan. Suno says v6 was developed with industry partners including Warner Music Group, BMG and Believe. The release adds editing that works on a finished song: change one section with plain-language instructions, build a mashup from several songs, and sample or isolate parts of a track. As v6 rolls out, Suno is retiring its previous models and moving entirely to the v6 generation.
Why: Suno was already our top pick for making a song from a prompt, and v6 is now the only Suno there is. The licensed-partner footing matters as much as the sound: it is the first Suno generation built with major rights holders, rather than in spite of them.
Alibaba's all-in-one video model: text, image, video and audio in, up to 30 seconds out
NewAdded Oct 19, 2026
Wan 3.0 is Alibaba's video generation model, in public beta from August 6, 2026 and formally launched August 24. One model handles text-to-video, image-to-video from a first frame or a first and last frame, and reference-based generation, taking text, image, video and audio as inputs. Clips run from 2 to 30 seconds at 30 fps, in 480p, 720p or 1080p. Alibaba Cloud Model Studio serves it as wan3.0-video, with a faster wan3.0-video-prime variant that Alibaba says matches its capabilities. Moving from Wan 2.7 needs only the model name changed, and the maximum length doubles from 15 to 30 seconds.
Why: Thirty seconds in a single generation, from mixed inputs, through one API, is a real step for teams already building on Model Studio. Rankings disagree on where it stands against Seedance and Kling, so it sits behind our picks for now rather than among them.
StepFun's 600B open-weight model for long-running agents, with a 1M-token context
NewAdded Oct 19, 2026
Step 5 Preview is StepFun's flagship model, published as downloadable weights on October 15, 2026 after an earlier hosted release. It is a sparse mixture-of-experts model with about 600 billion total parameters and 27 billion active per token, takes text and images as input, returns text, and supports up to one million tokens of context, aimed at coding and long professional work. Weights ship in BF16 and FP8 under the StepFun Model License 1.0, which StepFun describes as Apache-2.0-style with additional attribution and commercial-licence conditions. The model card gives example deployments of four B300 or eight H200 GPUs for FP8, and twice that for BF16, with vLLM, SGLang or Transformers.
Why: It is one of the largest open-weight models you can actually download, and it is built for the long agent runs that most open models still struggle with. The licence is not plain Apache 2.0 and the hardware bill is serious, so it is listed for teams that can run it, not as a default.
Nvidia's open vision-language model for agents, with open weights, data and recipes
NewAdded Oct 19, 2026
Nemotron 3.5 Super VL is Nvidia's open multimodal model, released October 15, 2026. It has 120 billion total parameters and 12 billion active, combines Mamba-2, mixture-of-experts and attention layers with a C-RADIO vision encoder, accepts images and text, returns text, and supports up to one million tokens of context. Weights ship in BF16 for customisation and NVFP4 for deployment, with a speculative-decoding companion checkpoint, under the OpenMDW-1.1 licence, which permits commercial use. Nvidia also published the datasets and training recipes. Artificial Analysis scores it 26 on its Intelligence Index, the highest of any US open-weight model, while using about 60% fewer output tokens per task than Nemotron 3 Super. A smaller sibling, Nemotron 3.5 Bolt (74B total, 9B active, 256K...
Why: It is the strongest open model from a US lab on independent measurement, and one of the few that publishes data and recipes as well as weights. Open models from China and the UAE still score well above it, so it ranks for openness and efficiency, not for raw capability.
A 24B open-weight reasoning model that runs only 2B per token, built for fast local agents
NewAdded Oct 19, 2026
LFM2.5-24B-A2B is Liquid AI's largest and final LFM2.5 model, released October 13, 2026. It is a mixture-of-experts reasoning model with 24 billion total parameters and 2 billion active per token, thinks before it answers, handles tool calls and multilingual text, and has a 131,072-token context, up from 32,768 in LFM2. Liquid reports 184 tokens per second on an Apple M5 Max and 114 on an AMD Ryzen AI Max+ 395. On its own tables it leads or matches Qwen3.6-35B-A3B on five of eight tool-use and agent benchmarks and all three instruction-following ones, while trailing larger models on coding and knowledge tests. Weights are on Hugging Face under Liquid AI's model licence, with support for llama.cpp, MLX, vLLM, SGLang and ONNX, and a hosted API on Liquid Console.
Why: Most open models good at tool use need a workstation GPU. This one runs fast on a laptop because only 2B parameters fire per token, which makes private, local agents practical. Liquid is candid that it is weaker at coding and knowledge, so pick it for agents that call tools, not for writing code.
Google's update to Nano Banana 2: better images and edits at half the price per image
NewAdded Oct 14, 2026
Nano Banana 2.1 is Google's October 6, 2026 update to Nano Banana 2 (Gemini 3.1 Flash Image), built on Gemini 3.6 Flash. It generates and edits images from text and reference images, at 1K, 2K and 4K, with multi-turn character consistency and optional grounding in Google Search. On the Gemini API it is gemini-nano-banana-2.1 at $30 per million image output tokens, which Google prices at about $0.034 per 1K image, $0.050 per 2K and $0.113 per 4K. It also runs in the Gemini app, Google AI Studio, AI Mode in Search, Flow and Stitch.
Why: It is a better model for less money. On the independent Artificial Analysis text-to-image arena it scores well above Nano Banana 2 and level with Grok Imagine Image 2.0 and MAI-Image-2.6, and Google's price per 1K image is about half Nano Banana 2's. Google lists it as the replacement for Nano Banana 2.
Alibaba's open-weight 7B image model, cut to 8 steps for fast generation and editing
NewAdded Oct 14, 2026
Qwen-Image-2.1-Turbo is an accelerated checkpoint of Alibaba Qwen's Qwen-Image-2.1, released on October 9, 2026. It uses the same 7B architecture but generates in 8 denoising steps, and handles both text-to-image and image editing from a reference, up to 2752 x 1536 at 16:9 and 2048 x 2048 square. The weights are on Hugging Face and ModelScope and load in Diffusers.
Why: It is one of the few strong image models you can download and run yourself, and the 8-step schedule makes it fast enough to iterate with locally. The licence is the catch: it is research and evaluation only, so it suits experiments and in-house prototypes rather than a product.
FreeBest Fast Open-Weight Image Model to Experiment WithVisit
Meta's first frontier model, for long-running agents, coding and computer use
NewAdded Oct 13, 2026
Muse Wonder is Meta Superintelligence Labs' October 12, 2026 model, a larger class than the Muse Spark line, released alongside a more capable Muse Wonder Full. Wonder is sold through the Meta Model API as muse-wonder-1.0 at $4 per million input tokens and $20 output ($0.20 cached input), with a one million token context window, and works with existing OpenAI SDK code by changing the base URL. It takes screenshots, charts, documents, images and video as input. Wonder Full is not on the API: it comes only with Muse Code's Power Usage and Maximum Usage plans. Both models will power Meta's Muse agent.
Why: It is Meta back at the frontier. At launch, VentureBeat reported, Artificial Analysis put Wonder a point behind GPT-6 Astra and Gemini 4 Argon on its Intelligence Index, and Wonder Full level with Claude Sonnet 5.5, behind only Claude Opus 5.5. Meta's own coding results are stronger still, and the API price matches Opus 5.5 per token.
Anthropic's small model at a tenth of Haiku 4.5's price, matching GPT-6 Luna
New this monthAdded Oct 8, 2026
Claude Haiku 5.5 is Anthropic's October 7, 2026 small model, replacing Haiku 4.5 as the recommended fast, cheap Claude. For prompts up to 100,000 tokens it costs $0.10 per million input tokens and $0.50 per million output, 90% below Haiku 4.5 and the same as OpenAI's GPT-6 Luna; above 100,000 tokens the price is $0.50 and $2.50. Cache reads are $0.01 per million. Anthropic says about 90% of Haiku requests fall under the 100,000-token line, so the average cost falls by around 75%. In Anthropic's tests it scores 1620 on GDPval-AA v2.1 against Luna's 1437, 72.4% on OSWorld 2.1 against 48.9%, and 39.2% on Terminal-Bench 4.0 against 16.4%. It is the first Haiku with adjustable effort levels, from low to max, and supports computer and browser use in beta through the SDKs. Output is text only.
Why: It resets what a small model costs and what it can do at that price: on Anthropic's tables it beats GPT-6 Luna on every shared benchmark at the same per-token rate, and the independent Intelligence Index puts its max-effort setting above Luna's too. For high-volume classification, extraction, routing and sub-agent work, it is the new default to test first.
Europe's 1-trillion-parameter model, in preview now with open weights promised
New this monthAdded Oct 7, 2026
Mistral Large 4 is Mistral's flagship, released in public preview on October 6, 2026. It is a natively multimodal mixture-of-experts model with 1 trillion total parameters and 49 billion active per token, trained on 3,800 NVIDIA Grace Blackwell GPUs in European datacenters across more than 160 languages. Mistral pitches it for cybersecurity, finance, legal and engineering work, and reports 61.7% on DeepSWE v1.1, 93% on Cybench, and results ahead of GPT-6 Astra on vals.ai's Finance Agent v2 and Harvey's legal agent benchmark. It is available now in Mistral Studio and Le Chat at $1.36 per million input tokens and $4.18 per million output tokens. Mistral says open weights follow at the end of October, after safety testing with trusted partners and governments.
Why: It is the largest model Europe has shipped and the first Mistral release in a year that competes on capability rather than only on licence and price. The price is a fraction of the US frontier, and if the weights land as promised it becomes one of the largest open-weight models outside China. It is listed on Mistral's own benchmark claims, clearly labelled, with the independent Intelligence Index score shown alongside.
A 501B open-weight coding and agent model from a US lab, under Apache 2.0
New this monthAdded Oct 7, 2026
Beam is the first model from Reflection AI, the Nvidia-backed New York lab founded by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. Announced October 5, 2026, it is a sparse mixture-of-experts model with 501 billion total parameters and 23 billion active, a 1-million-token context, and text-only input and output. Reflection reports 44.4% on DeepSWE v1.1, 80.1% on Terminal Bench v2.1, 80.9% on SWE-bench Verified and 90.5% on GPQA Diamond, broadly level with Z.ai's GLM-5.2 and behind Qwen3.8 Max and Kimi K3, and claims it reaches GLM-5.2's scores with 3 to 4 times less inference compute. The weights went up on Hugging Face on October 16 under Apache 2.0, in BF16, FP8 and NVFP4, with day-0 support in Transformers, vLLM and SGLang, a Rust-based agent harness, and...
Why: The US has had few serious open-weight models since Llama stalled, and Beam is a credible one under a fully permissive licence. It does not lead its Chinese peers, and Reflection's own tables say so. It is listed for the efficiency claim and the Apache 2.0 licence, ranked below the open models it trails until it is independently tested.