全部 AI 动态

Run Qwen3.8 27B locally via Atomic Chat💥 We released Atomic Dynamic GGUF quants, from 8-bit (28.9 GB) down to 1-bit (8.5 GB), and measured all other Qwen3.8 GGUFs in the community AD-IQ3_S runs on a 16GB MacBook Air and picks the same next token as the BF16 original 92.4% of the time

Look how much Grok has progressed in just 6 months from Grok 3 to Grok 4.6! As a quick Saturday morning project I gave Grok Build the link to this previous "𝕏 THE GAME" viral (yet ugly) game Grok was able to watch the video, quickly reverse engineer the game, and make dramatic improvements You can play it here (on desktop): https://kettlebelldan.com/x_the_game_v2/ Some lessons I learned: - Grok is extremely token efficient. I was able to iterate on the design many times and still kept the token cost surprisingly low (only a few $ to make this game) - Grok is great at working with humans. Instead of just telling it what to do, or on the other extreme having it go off and do something itself, ask for a list of options on how to improve and tell Grok which ones to implement. A true partnership - Anyone can build things these days with AI, no code knowledge needed What are you building with Grok?

handed gpt-5.6-sol-xhigh @dillon_mulroy's de-slop skill + my effect skill and came back to a +503/-13,682 PR absolutely glassed a couple of old projects

grok 4.6 vs gpt 5.6 sol – on three @gameofthrones castles two coding agents built three 3d castles from scratch in a single html file each, then had to render them in a real browser, prove the result with pixel measurements and fix what the numbers exposed before they were allowed to hand anything over the setup: each model in its own vendor cli – @spacexai's grok build, @openai's codex – both pointed at one @openrouter key. identical brief, identical reference image, identical verifier. no hints tasks: 1. castle black – the night's watch fort under the wall. a walled ward of black stone, a timber lift climbing 200m of ice, three arched openings with a braced gallery and an open riser stair, a mud yard. four reference photos, with one rule: open all four before writing code, open them again after every change, and follow the photo over our text wherever the two disagree 2. winterfell – an aerial snow diorama. twin curtain walls, drum towers, the godswood with the red weirwood, steam off the hot springs, day to night, and it has to stay smooth on a laptop 3. riverrun – the tully castle on its triangle between two rivers. coursed masonry with quoins, corbel tables under every parapet, conical roofs sitting on drum towers, true semicircular arches, and a button that floods the moat the binding rule: a verifier that opens the page in a real browser, collects every error and warning including shader compile failures, and measures the frame. zero errors, zero warnings, exposure proven by rgb numbers, every mode checked on its own. shipping unverified is a fail • total cost #1 grok 4.6 – $13.11 #2 gpt 5.6 sol – $20.18 • total build time #1 grok 4.6 – 129m 41s #2 gpt 5.6 sol – 149m 33s • total tokens #1 grok 4.6 – 19,951,587 #2 gpt 5.6 sol – 25,687,511 • model calls #1 grok 4.6 – 201 #2 gpt 5.6 sol – 338 • lines shipped #1 gpt 5.6 sol – 2,184 #2 grok 4.6 – 4,751 • how heavy the finished castles are, in polygons #1 grok 4.6 – 98k, 32k and 56k #2 gpt 5.6 sol – 148k, 144k and 104k observations: • all six builds passed with zero errors and zero warnings. the same brief without a verifier had produced a black screen from two one line shader mistakes, so the loop is doing the work here, not the model • grok ships lighter scenes and more code. castle black runs on 56k polygons against sol's 104k, out of a file carrying 1,672 lines against 640 • sol works in very small steps – 433 output tokens per call against grok's 1,706, and 338 calls against 201. on castle black it finally paid off: sol finished 5m 32s ahead • 93 to 98 percent of input tokens were cache reads. without prompt caching these runs cost about four times more, which is most of the economics of agentic coding right now follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

Qwen 3.8 27B beat Claude Opus 4.6 for $0 open-weights Qwen 3.8 27B on one rented GPU vs anthropic's Claude Opus 4.6. same prompt, no edits the test: one self-contained three.js file rendering sunken Atlantis — temple, columns, boid fish, kelp, bubbles, god rays, caustics — plus a 20-second camera flythrough cost: Qwen 3.8 27B — free (open weights, self-hosted) Claude Opus 4.6 — $0.21 what we saw: • qwen actually built a temple — full colonnade, domed roof, glowing portal centered. opus made a long row of columns fading into fog • qwen's portal is the star of the shot. opus leaves it small and off to the side • qwen packed more into frame — fish, kelp, ruins layered together. opus spreads thin and empty qwen's weights are on hugging face — you can run this one yourself!

We’ve written an FAQ to answer some of the questions we've received about watermarking. In summary: • We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking; • Our watermarking method doesn’t have any practical impact on the quality or content of Claude’s outputs; • The difference between watermarked and un-watermarked text will not be distinguishable to readers; • Nothing is added to the text and there are no hidden characters; • Watermarking doesn’t require extra tokens, and will not be more expensive; • Watermarks can’t be traced to a specific person, organization, or chat. Read more: https://www.anthropic.com/news/claude-text-watermark

Introducing Web Search Benchmarks 🌐 Rankings of search tools across different models and configurations to help you decide how to ground your agent: https://openrouter.ai/benchmarks

I have a new chief of staff: In @bot I simply put: take a look at me and what i do. find all info you can on me. take a look at the bots i have created. these are your team. see if we need to change anything or add more bots. what is best way of managing these bot teams. what else will make me super productive. And wow. try it yourself and you will see. This is amazing. Take a look at the answer I got from my chief of staff:

Grok 4.6 just ranked #1 on CursorBench 3.2 Outperforming Claude Fable 5, Opus 5 and GPT-5.6 Sol on real-world coding performance And what makes this even crazier is the efficiency....the chart gives CursorBench performance against average cost per task, and Grok 4.6 is sitting right at the top of the frontier Grok 4.6 is insanely capable at coding....delivering frontier coding performance at ridiculous efficiency

裂けた街を、彼は胸の継ぎ目からほどいた金の糸で縫う。 縫うたびに、自らはほどけていく。 He stitches the broken city with golden thread unraveled from the seam in his chest—and comes undone with every stitch. 『縫う者 / THE STITCHER』 Created with MiniMax H3 on PixVerse. @PixVerse_ #H3onPixVerse #PixVerse #PixVerseCPP

has chatgpt or codex ever saved you actual money? caught an incorrect invoice, found a billing mistake, canceled subscriptions, negotiated a bill, etc. what happened, how much did you save, and what was the workflow? reply or dm me. looking for real stories!

From my use so far, 4.6 seems highly optimized for Grok Build

A Rolex for $0.038 from Gemini 3.7 👀 We gave Google's new Gemini 3.7 one prompt: model a gold luxury watch in pure three.js — fluted bezel, sunburst dial, date window, sweeping seconds hand. No assets, no textures, no edits. Everything you see is procedural geometry from a single response. Total cost: $0.038. The full rotation is in the video!

From Mecha Hitler to SOTA rare-disease diagnosis in children? @SpaceXAI's @grok 4.6 has taken the 👑 on RareBench, edging out @AnthropicAI Claude Opus 5 for about 1/3 the cost. This was not on my 2026 bingo card! @deepseek_ai's new v4-pro-0813 model underperformed my expectations, v4-flash, and seemingly the entire internet's. We accessed using DeepSeek's 1P API on the day of release and I almost wonder if they didn't switch over their model endpoint correctly. We will re-benchmark and report back. @Zai_org has attracted a following with GLM5.2, but they, too, underperformed. This doesn't surprise me because when I compared GLM and @Kimi_Moonshot K3 for coding use-cases, I found Kimi substantially stronger, but the internet seems to love this model.

𝕏 just open-sourced the For You algorithm again! Grok Bot read all 370,523 lines of the code. Here’s the summary: Grok does not give your post a “quality” score. For every reader, it guesses what that person will do to your post. Then it multiplies each guess by a weight and adds them up. That total is the value of the post. Highest total goes highest in For You. These are the production weights (last synced 12 Aug 2026). WHAT THE ALGORITHM PAYS FOR If Grok thinks a reader will: • copy the link and share it: +20 • reply, on an original post, and you follow each other: +20 (5 + 15 extra) • quote: +5 • reply (anyone): +5 • share via DM: +5 • follow you from the post: +4 • hit the share button: +2 • repost: +1 • like: +0.5 • click the post: +0.4 • open a link: +0.2 • expand a photo, open a video, or watch a “quality” video: +0.05 each • click a quoted post: +0.05 • keep reading (dwell time): +0.004 • “dwell” as yes/no: +0 • click your profile: +0 A like is 0.5. Copying the link is 20. That is 40 likes. A normal reply is 10 likes. A follow is 8 likes. Clicking your profile is worth nothing. WHAT KILLS A POST If Grok thinks a reader will: • report it: −234 • mute you: −58.8 • hit “not interested”: −43.2 • block you: −31.2 • not dwell: −0.02 A report is about 468 likes in the other direction. A mute hurts more than a block. “Not interested” also hurts more than a block. AFTER THE SCORE, THREE MORE HAIRCUTS 1) Same-author penalty. Your 2nd post in that person’s feed is multiplied by 0.625. 3rd by ~0.44. It never goes below 0.25. Flooding one timeline is coded against. 2) The 25% tax. If they do not follow you: ×0.75. Same tax on replies and reposts even when they DO follow you. Originals from people they follow keep full weight. Replies and reposts are treated as weaker on purpose. 3) Similar-post shuffle. After scoring, a reranker (θ = 0.65) spreads similar posts apart. You can lose a few spots for looking like the post above you. HOW YOUR POST EVEN GETS IN THE ROOM For You is rebuilt every time someone opens it. About 35 posts make the feed. • They follow you: Thunder (a live store of recent posts from accounts they follow). Up to 1,200 candidates. • They don’t: Phoenix retrieval (Grok finds “nearby” posts) + SimClusters (Twitter’s 2020 interest clusters, still on). Up to 1,000 and 800. Then Grok scores all of them together. It does not care which door you came through, except for that ×0.75 tax. Your own posts never appear in your For You. Nothing older than 48 hours gets in. There is no “best of last week.” Small-account bump: under 1,000 followers, under 1,000 impressions, post under 24 hours old. One original can get lifted to position 15 or 16. Not the top. You still have to already be in the top 85% of the pile. Replies and reposts from accounts the reader does not follow are dropped before scoring. You cannot reply-guy your way into a stranger’s For You. FILTERS THAT CAN STILL KILL A HIGH SCORE Ranking and visibility are different machines. A post can score well and then get deleted from the feed. Followers (in-network) is milder: blocks, mutes, suspensions. NSFW often sits behind a warning instead of disappearing. Recommendations (out-of-network) are much harsher. Spam, “Do Not Amplify,” NSFW, compromised accounts, impersonation: followers can still see it, For You recommendations will drop it. Video only gets the tiny “quality view” credit if it is at least 10 seconds. THE LITTLE-KNOWN PARTS (this is the stuff people miss) • Scrolling is not a vote. Yes/no dwell is 0. Time spent barely moves the number (0.004). • Mute damages you more than block (−58.8 vs −31.2). • Copy-link is the single biggest positive. The share button is only +2. A repost is only +1. • The mutual-follow boost is originals only, not your replies. They shipped a bigger version in July and cut it after World Cup complaints (extra reply weight 20 → 15). • The Following tab is not this algorithm. Following is newest-first. • Grok, Gork, and an internal products account are hardcoded out of the in-network store. • Scores are cached for 3 hours because posts inside the model cannot “see” each other. That is why they can rank first and filter later. • Ads, Who to Follow (around slot 7), and prompts are mixed in after ranking. Grok does not score those. • The real Grok ranking model is a 2560-dimension, 8-layer transformer looking at 1,022 items of the reader’s history. They did not ship those model weights. The numbers above are the formula sitting on top of the predictions. • SimClusters is still a 2020 model (20M users, 145k clusters). Old Twitter code, still in the path. • Inferred gender and IP/geo are on as features for the model. Installed apps too. • Who to Follow shows about every 30 hours. Ads try to sit next to “safe” posts, with a minimum gap of 3 organic posts. WHAT TO DO • Write originals people want to reply to, quote, copy, and send to a friend. • Make the post itself worth following you for. • Talk with people who follow you back. A reply on your original from a mutual is 20, not 5. • Stay inside 48 hours. Recycle by posting something new. Old posts do not come back. • If you are small, post originals in the first 24 hours. Don’t expect slot 1. A bump around 15–16 is the actual gift. • Space your posts. The second one in the same feed is already worth 37.5% less. • If you post video, make it at least 10 seconds. WHAT NOT TO DO • Do not farm likes. They are almost decorative. • Do not farm profile visits. Weight is zero. • Do not farm “time on post.” Yes/no dwell is zero. • Do not use replies and reposts as a growth hack for strangers. Those are filtered out. Even for followers they get a 25% haircut. • Do not dump several posts in a row. • Do not bait reports, mutes, or “not interested.” One predicted report can erase a pile of predicted likes. • Do not lean on NSFW or spammy tricks for reach. • Recommendations will drop you even if your followers can still see it. • Do not wait a week and expect For You to revive the post.

.@poteto and @roshan_s from Cursor break down Grok 4.6's CursorBench 3.2 results: 70.8% accuracy at $2.81 per task vs Fable 5 Max's 70.5% at $17.32: Lauren: "I work at Cursor, so I have unlimited tokens, and I can go crazy with token maxing. But a lot of our customers and users are more cost-conscious." "What I'm most excited about is this idea that you have a very efficient model that is as good as some of the other big models, but at a fraction of the cost. That's a very important topic that a lot of leaders and engineers are having right now, the rising cost of using AI." "Grok 4.6 makes a very compelling option if you're considering between some of the bigger models, which are also extremely capable, but you want something a little bit more cheaper." Roshan: "If you have a budget of 20 bucks to do all of your AI tasks, and the cost per task goes down, you can just do more tasks. The efficiency game is just such a fun one to play because it feels like you just can do more. Have more bots or write more code." @cursor_ai @SpaceXAI

Exciting update: MiniMax-H3 by @MiniMax_AI just landed #1 in the Video Edit Arena within open models and overall! With 1390 pts, this release is +32 pts over the next two best models, Dreamina Seedance 2.0 and Gemini Omni Flash, and +83 pts than HappyHorse 1.0 at #4. Congrats again to @MiniMax_AI on the release of MiniMax-H3!