全部 AI 动态

Primer tops BigFinanceBench. We ran @primerapp_ on @RogoAI's BigFinanceBench. Primer tops the leaderboard: 79.1% final-answer accuracy vs 55.8% for the best frontier model. BigFinanceBench is the real deal conceptually in terms of a benchmark: 928 questions of actual analyst work, graded on the working as well as a final answer. It's now the only finance eval on @OpenAI GPT-5.6 launch page. This is such a great demonstration of the power of an agent harness. At time of testing, Primer's agent was powered by GPT-5.5 which when tested 'naked', scored 55.8% but inside Primer's harness: 79.1%. As we saw from the BigFinanceBench scores of Fable, Opus 4.8 and GPT-5.6, next generational models are not closing this gap in finance, which is why we are laser focused on improving our harness to deliver the highest performing finance agent globally. While verifying our results and understanding our errors, we found errors in 12% of the benchmark's questions, which we have detailed in our blog post here: https://www.primerapp.com/blog/primer-tops-bigfinancebench/

Totally jamming with Codex voice mode. Talk in one chat, it's where you riffa and when you land on a solid outcome it will fire up a new session to start the work using all of your existing tools. The voice chat is your manager that can check in on all other chats. It's sick.

qwen 3.8 max vs deepseek v4 flash 0731 vs kimi k3 vs gpt 5.6 sol – on rubik's cube and chess four frontier models built a rubik's cube stand and solved it, then built a chess board and played claude opus 5 on it the setup: @nousresearch's hermes agent cli on @openrouter tasks: 1. cube – build a 3d rubik's cube with a cli and a @threejs viewer, then solve an identical scrambled position on your own stand 2. chess – build a 3d chess stand, then play white against claude opus 5 as black, live, one move at a time. no engine, no solver, no opening book on either side. stockfish depth 14 grades every chess ply afterwards; neither player sees the score models: @deepseek_ai v4 flash 0731, @openai gpt-5.6 sol, @kimi_moonshot kimi k3, @alibaba_qwen qwen 3.8 max gpt-5.6 sol and deepseek v4 flash solved their cubes – sol in 24 moves and seventeen seconds, deepseek in 32. qwen and kimi never got there, giving up at 96 and 207 moves then all four built chess stands and played white against claude opus 5 on them, and all four resigned: deepseek on move 13, sol on 19, kimi on 21, qwen holding out longest at 29 - build time, both stands #1 gpt-5.6 sol – 16m 43s #2 deepseek v4 flash – 97m 39s #3 kimi k3 – 166m 09s #4 qwen 3.8 max – 215m 08s - build attempts before a working stand #1 gpt-5.6 sol – 3 #2 qwen 3.8 max – 4 #3 kimi k3 – 4 #4 deepseek v4 flash – 5 - total tokens #1 gpt-5.6 sol – 6,713,754 #2 qwen 3.8 max – 17,272,507 #3 kimi k3 – 22,427,504 #4 deepseek v4 flash – 27,417,442 - total price #1 deepseek v4 flash – $0.557 #2 gpt-5.6 sol – $6.319 #3 qwen 3.8 max – $10.270 #4 kimi k3 – $16.667 observations: • deepseek v4 flash is the cheapest model here by a margin nobody else is near, and it got there while being the least efficient of the four. it burned 27.4m tokens – more than anyone, 5m more than kimi – and still finished both benchmarks for $0.557. that is $0.02 per million tokens against kimi's $0.74. it also needed the most passes to produce working stands, five, and that did not matter: all five deepseek passes together cost a thirtieth of kimi's two • so what deepseek cannot do is get it right the first time. what it can do is get it right the fifth time, for half a dollar. that is a different thing to be buying – not a good first draft, but the option to keep asking • gpt-5.6 sol is the opposite profile and the strongest of the four on pure efficiency. 16m 43s to build both stands, 6.7m tokens, three passes – under 40% of the next lowest token count and a quarter of deepseek's, on an eighth of qwen's clock. it also solved the cube fastest of anyone, 24 moves in seventeen seconds. sol is what you reach for when you want the answer now and can absorb $0.94 per million • sol's weakness is in what it does not check. its chess viewer deleted the capturing piece instead of the captured one, so pieces disappeared off the board mid-game – a defect the fifty-cent deepseek stand did not have. fast and terse turns out to be the same dial as fast and unverified • qwen 3.8 max is not the cheap open-weights option it gets treated as. $10.270 across the two benchmarks, second most expensive of the four, 18x deepseek, and by a distance the slowest – 215 minutes of build time, nearly thirteen times sol's. what the money buys is judgment: it played eighteen moves without a single error worth a hundredth of a pawn, then made exactly one bad move in the whole game, and averaged 44.6 centipawns lost across the longest game any of the four managed. it also could not solve a rubik's cube in 96 tries • kimi k3 is the one line with no reading that flatters it. most expensive at $16.667, last on the cube at 207 moves, last at chess at 478 centipawns lost per move. it is also the model that verified hardest – on the cube it wrote its own integrity check instead of trusting its output. that makes the result worse rather than better: the checking was real, and the reasoning underneath it still was not follow @thehypedotnews for 24/7 ai news, analysis and breakdowns

Google's chief scientist is leaving after 27 years to build AI that can run its own research cycle with little human help. Jeff Dean was employee 30, and he wrote the storage and computing systems that showed the rest of the industry how to serve a global audience. He also started Google Brain in 2011 and drove the TPU units. His new firm is called Discovery Loop, after the cycle it wants to automate, which is hypothesis, then experiment, then evaluation of the result. Dean says whole domains can be computerized that way, so you get more experiments and better ones out of the same year. It begins by automating machine learning research itself, before moving toward hardware design, drug discovery and clean energy. Joining him are Sanjay Ghemawat, his collaborator of two decades, plus Quoc Le and Oriol Vinyals from the Brain and Gemini years. Radical Ventures and Khosla Ventures are co-leading a seed round that is not yet closed, and Alphabet came in as founding investor and cloud partner. Alphabet will supply their computing power for at least a year, so the four keep frontier-scale hardware without paying to build it. Google shares fell about 4% on the news.

SILENT SKY. a narrative video to tell a story. a forest disappeared. the silence stayed. made in PixVerse. @PixVerse_ #pixverse #pixversecpp

Big news: Opus 5 (Max) is now #1 in the Fullstack Code Arena with 1,699 points! The Fullstack Leaderboard shows overall rankings across AI models on full-stack web development tasks: multi-step reasoning, tool use, and end-to-end app generation. Congrats again to the @AnthropicAI team on Opus 5 (Max)!

Exciting news: Qwen3.8-Max by @Alibaba_Qwen is #2 in Image-to-WebDev Arena! With 1,631 pts, it’s trailing only Claude Opus 5 (Max) by 39 pts! Congrats again to @Alibaba_Qwen on this huge release!

Van Gogh city builder game.

I set up Gemini Spark to write my own weekly AI news digest for me, and I'm still stuck on how little effort I had to put in. This is the exact prompt I gave it, just once: "Every Tuesday at 12:15 PM, create a new Doc in my Drive, find the 5 biggest AI stories from that week, and format each one properly." And here's what it actually produces, every week: 1. A new Doc, created and named on its own 2. The 5 biggest AI stories from that week, found and picked automatically 3. A bold headline and a summary for each one 4. The key impact spelled out, not just the news 5. A source link on every single story This runs on a schedule so I never have to think about it again. I didn't ask anything on Tuesday, I just opened the drive and the news digest doc was already there. Just like this, you can use Spark to automate any recurring task.

My honest review of Hailuo 3 after testing it on PixVerse... One thing that genuinely surprised me was how well it handled scenes that are often difficult for AI video models. It was able to generate clips that other models might flag or heavily restrict, while still producing cinematic, coherent results. For this project, I used a storyboard-first workflow and let Minimax Hailuo 3 (MiniMax H3) on @PixVerse_ bring each shot to life. Some standout features: 🎥 Up to 15-second clips in 2K resolution 🖼️ Up to 9 reference images in a single generation 🎨 Style reference controls 🧠 Strong physical understanding for complex action and camera movement ✍️ Better prompt adherence and more accurate text generation 🎙️ Built-in lip sync 🌍 Support for multiple minority languages What impressed me most wasn't just the image quality—it was how well the model followed a structured storyboard made with GPT Image 2 and delivered cinematic camera moves, consistent characters, and polished action sequences. If you're building AI short films or commercials, this is definitely a model worth experimenting with. #PixVerseCPP

お疲れ様です♪ @PixVerse_ さんでランチですよ✨️ #pixversecpp PixVerse CPP Saumon en papillote (サーモン・アン・パピヨット)でございます❤ 紙包み焼きです♪ こんな分厚いサーモン、いただいてみたいですね(〃艸〃)✨️ 午後も楽しく頑張りましょうです٩(ˊᗜˋ*)و♪

GPT-Live 可以边说边听。 为了让这在 ChatGPT 规模下显得自然，我们从头重建了从客户端到模型的语音技术栈。 这一新架构让音频持续流动，因此更深入的推理和工具调用不会打断对话。

This is an interesting admission from Claude, and it's consistent with my experience with it. It gets more stubborn and attached as the context window gets longer. Kind of like people as they age, arguably.

yeah you hate opus slop now but in a month you're gonna be so sick of simplified technical english i guarantee it