Ideogram was founded by former Google Brain researchers who worked on Imagen, Google’s own text-to-image system. When that team releases an open-weight model, you pay attention.
Ideogram 4 tops the open-weight design leaderboard by a margin that isn’t close. Professional designers picked it first in blind typography tests nearly half the time. At 9.3B parameters it beats open models three times its size on text rendering.
brew-browser: Native Homebrew GUI for macOS to Manage Packages, Services & Brewfiles
https://firethering.com/brew-browser-homebrew-gui-for-macos/
#opensource #brew #developers #developertools #macos #homebrew
brew-browser doesn't replace Homebrew. It simply gives it a proper macOS interface. You can see what's installed, search thousands of packages, upgrade software, manage services, and create Brewfile snapshots without memorizing commands or keeping multiple terminal windows open. Everything still runs through the real brew CLI, so you're not learning a new package manager or dealing with a compatibility layer. It can pull information into one place. Installed packages, available updates, storage usage, trending Homebrew packages, services, snapshots, and even optional vulnerability scanning all live inside a native desktop app that feels at home on macOS.
Google Built Gemma 4 12B Without Multimodal Encoders
https://firethering.com/google-built-gemma-4-12b-without-multimodal-encoders/
Every multimodal model you've used has the same basic system. Text goes in one way, images go through a vision encoder first, audio goes through an audio encoder first, and then everything gets handed off to the language model in a form it can work with. The encoders are load-bearing and you don't just remove them.Google actually removed them.Gemma 4 12B takes raw image patches and raw audio waveforms and projects them directly into the same embedding space as text tokens. There is no vision encoder or audio encoder. One decoder handling everything.
Donut Browser: Open Source Anti-Detect Browser With Unlimited Isolated Profiles
https://firethering.com/donut-browser-open-source-anti-detect-browser/
If you've ever managed multiple accounts in the same browser, you've probably run into the usual mess. Wrong account logged in. Cookies bleeding between sessions and more. Donut solves that by treating every profile as its own browser. Create a profile, attach a proxy if you want, and it gets its own cookies, storage, extensions, fingerprint, and network settings. Open five profiles and it feels like you're using five separate browsers. Everything stays local. There's no account to create. You download the app, create profiles, and get on with your work. If you're managing client accounts, testing websites, keeping work and personal browsing apart, or building automation workflows through the local API, Donut gives you a clean way to keep identities separated without turning your browser into a headache.
Most models quit around submission 30 because they stop finding improvement and exit on their own. That's what happened when MiniMax ran a CUDA kernel optimization task against a field of frontier models. Every model except two called it done within the first 30 submissions. M3's best result came on submission 145. After 24 hours. After multiple plateaus where the numbers stopped moving and a reasonable model would have concluded there was nothing left to find. That's the thing MiniMax released yesterday. An AI model with a 1M token context window, native multimodality, and apparently a problem with knowing when to stop.
Anthropic Files for an IPO. AI Is Entering Its Public Company Era.
https://firethering.com/anthropic-ipo-ai-public-company-era/
#anthropic #darioamodie #claude #claudecode #news #ai #technews
Anthropic has officially taken its first step toward becoming a public company. In a brief announcement on Monday, the company said it had confidentially submitted a draft S-1 registration statement to the U.S. Securities and Exchange Commission for a proposed initial public offering. The filing doesn't reveal a share price, a fundraising target, or even a timeline. For now, it simply gives Anthropic the option to go public once the SEC review process is complete. Just a few years ago, Anthropic was a small group of former OpenAI researchers trying to build an alternative vision for advanced AI. Today, it sits among the handful of companies shaping the industry's future and that's why this filing matters. It's one of the world's most influential AI labs beginning the transition from a privately funded research company to a business that may eventually answer to public shareholders. For most of the AI boom, the biggest bets were made behind closed doors. Venture firms, sovereign wealth funds, and tech giants supplied the capital while the public watched from the outside. Anthropic's filing suggests that era may be starting to change.
OpenAI says one of its internal reasoning models has solved a math problem that has been there on mathematicians’ desks since 1946.
The problem, first posed by legendary mathematician Paul Erdős, looks almost absurdly simple. Given a set of points on a flat plane, how many pairs can be exactly one unit apart? People have spent nearly 80 years trying to pin down the answer.
https://firethering.com/openai-ai-solves-80-year-old-math-problem/
#openai #maths #ai #news #technews #trending #gpt #chatgpt
OpenAI says one of its internal reasoning models has solved a math problem that has been there on mathematicians' desks since 1946. The problem, first posed by legendary mathematician Paul Erdős, looks almost absurdly simple. Given a set of points on a flat plane, how many pairs can be exactly one unit apart? People have spent nearly 80 years trying to pin down the answer. OpenAI's model didn't just make progress on the problem. According to the company, it disproved a longstanding conjecture that many researchers believed was essentially correct.
MiniCPM Desk Pet turns the MiniCPM AI model into a desktop companion that lives alongside your workflow. Install the app, follow the setup wizard, and within a few minutes you can chat with a local AI pet directly from a floating desktop bubble.
Conversations run on your machine using the local model. The pet can stay visible while you work, react to activity from tools like Cursor, Claude Code, and Codex.
https://firethering.com/minicpm-desk-pet-open-source-local-ai-desktop-pet/
#minicpm #coding #opensource #ai #claudecode #developers
MiniCPM Desk Pet turns the MiniCPM model into a desktop companion that lives alongside your workflow. Install the app, follow the setup wizard, and within a few minutes you can chat with a local AI pet directly from a floating desktop bubble. The app checks your environment, downloads the model, warms it up, and simplify the complexity of the setup Once everything is ready, conversations run on your machine using the local model. The pet can stay visible while you work, react to activity from tools like Cursor, Claude Code, and Codex, and even take on different personalities through character adapters. It's part local AI assistant, part desktop pet.
MiniCPM5-1B Shows Why the Small-Model Race Isn’t Over
https://firethering.com/minicpm5-1b-small-model-reasoning/
#ai #llm #minicpm #genai #opensource #openweights #technews #tech
A 1B model scoring 40.42 on AIME 2025 should not be possible. AIME is the American Invitational Mathematics Examination, the kind of test that filters out most humans who attempt it. Qwen3-0.6B scores 16.25 on the same benchmark. LFM2.5-1.2B, a larger model, scores 31.88. MiniCPM5-1B, at roughly one billion parameters, beats both. OpenBMB just dropped MiniCPM5-1B, the first model in their MiniCPM5 series, and it's built specifically for the scenarios like on-device deployment, resource-constrained environments, local inference on consumer hardware. The AIME score is surprising. The telecom agent benchmark is even more surprising. And then there's the desktop pet. We'll get to that.
StepFun Says Step 3.7 Flash Matches 97% of Claude Opus 4.6’s Coding Performance at One-Ninth the Cost
https://firethering.com/stepfun-step-3-7-flash-agentic-coding-cost-efficiency/
#stepfun #claude #opus #coding #ai #technews #tech #llm #opensource
$0.19 vs $1.76. That's the per-task cost of running Step 3.7 Flash with Advisor Mode enabled versus Claude Opus 4.6 on SWE-Bench Verified. The Flash model scores 76.3% to Opus 4.6's 78.7%. Two percentage points of difference. Nine times cheaper to get there. For anyone building agentic coding workflows at scale that math changes the decision about which model actually belongs in production. Frontier performance has been getting cheaper for a while but this is a specific, benchmarked claim with a specific cost figure attached.