cedric
@cedric_chee
Followers
2K
Following
6K
Media
923
Statuses
8K
SWE, Logorhythms | @fastdotai alumni, Independent LLM researcher | Code: https://t.co/KwpQnH7gRo | ex-entrepreneur @AntlerGlobal | genuinely curious
The Matrix
Joined November 2007
🧵1/3 Claude Code and Qwen3-Coder is so good at multi-tool use and selection, reliably! resembling Claude Sonnet 4. It created this Minecraft clone in a few prompts. Best of all, it doesn't require a Claude MAX plan.
2
2
29
Gemini 2.5 Flash Image model is now called Nano Banana.
0
0
3
Packed with tricks. Despite being a non-reasoning model with a compact parameter count, it delivers highly competitive performance among leading models, with exceptional strengths in agentic tool use, general reasoning & coding. 3 key innovations drive it:.1) ScMoE with
The technical report of @Meituan_LongCat LongCat-Flash is crazy good and full of novelty. The model is a 560B passive ~27B active MoE with adaptive number of active parameters depending on the context thanks to the Zero-Computational expert. 1) New architecture.> Layers have 2
0
0
5
xAI is cooking so hard! They really want to beat Claude. The post mentioned:.> A new variant that supports multimodal inputs, parallel tool calling, and extended context length is already in training. > improvements arriving in days rather than weeks.
0
0
0
grok-code-fast-1 is an efficient model, up to 4x the speed and 1/10th the cost of other leading agentic models. Prompting guide: use it for agentic-style tasks rather than one-shot queries.
Introducing Grok Code Fast 1, a speedy and economical reasoning model that excels at agentic coding. Now available for free on GitHub Copilot, Cursor, Cline, Kilo Code, Roo Code, opencode, and Windsurf.
1
0
4
I broke ChatGPT-5 Canvas and pushed Codex CLI to its limit 2 weeks ago to perfect a high-performance ray-tracing simulator. The latest Codex CLI revamp is still poor. I hacked session resume to exceed 150+ multi-turn chats and >500k (80%) token context. Don't underestimate
0
0
0
Now you can add OpenAI's coding agent, Codex as a panel in VS Code. Works with ChatGPT plan (Plus, Pro, Team, Edu, and Enterprise)
We’re releasing new Codex features to make it a more effective coding collaborator:. - A new IDE extension.- Easily move tasks between the cloud and your local environment.- Code reviews in GitHub.- Revamped Codex CLI. Powered by GPT-5 and available through your ChatGPT plan.
0
0
6
CLI agents are simple and portable. Zed runs Gemini CLI via ACP, bringing terminal-level power directly into the IDE. Cool!
🚀 New in Zed: Bring your own agent, starting with @Google’s Gemini CLI. → Try Gemini CLI with full code context in Zed. → Build & run multiple agents in your editor. → Powered by the new Agent Client Protocol (ACP). Learn more:
0
0
0
Sonic was Grok Code. It's really fast!.The price is way better than I expected. $0.2/M input, $1.5/M output toks. Good for medium complexity problems. GPT-5 for very complex problems.
Fifth model launch of the day‼️Grok Code Fast, from @xAI . At 92 tokens per second so far, it's one of the fastest coding models on OpenRouter, and comes with 256k context
2
2
6
Anthropic released Claude Chrome extension as a limited research preview. If they make this work with Claude Code, it will be phenomenal. They reduced the attack success rate of 23.6% to 11.2% when safety mitigations were added.
We’ve developed Claude for Chrome, where Claude works directly in your browser and takes actions on your behalf. We’re releasing it at first as a research preview to 1,000 users, so we can gather real-world insights on how it’s used.
0
0
2
WAN cinematic speech-to-video model dropped. Wow!.
🚀Introducing Wan2.2-S2V — a 14B parameter model designed for film-grade, audio-driven human animation. 🎬Going beyond basic talking heads to deliver professional-level quality for film, TV, and digital content. And it’s open-source!. ✨ Key features:.🔹 Long-video dynamic
0
0
0
Impressive!. Prompt: Recreate this dog as a video game character, and place the character in a level of a 3D voxel or blocks world MineCraft video game.
0
0
0
Very impressive nano-banana model image editing benchmarks! I am so pumped to give it a spin.
Introducing Gemini 2.5 Flash Image (aka nano-banana), our SOTA image generation and editing model 🍌. As you might have already seen, this model excels at character consistency, creative edits, and has Gemini's world knowledge!
1
0
2
🍌 is here! It's Gemini 2.5 Flash with editing and image generation.
Image generation with Gemini just got a bananas upgrade and is the new state-of-the-art image generation and editing model. 🤯. From photorealistic masterpieces to mind-bending fantasy worlds, you can now natively produce, edit and refine visuals with new levels of reasoning,
0
0
1
RT @cline: There are 3 pervasive patterns in coding agent development that we deliberately avoid at Cline:. 1. Multi-agent orchestration.2.….
0
30
0
RT @karpathy: Continuing the journey of optimal LLM-assisted coding experience. In particular, I find that instead of narrowing in on a per….
0
997
0
Grok-2 model spec decoded directly from the HF repo (since the card isn’t up yet):. - Parameters: On disk the checkpoint is ~500GB of bf16 weights; that’s roughly 250–268B. Only a fraction is active per token due to MoE (see below). - Architecture: Residual MoE Transformer. -.
0
0
2