Schwarzer Ritter

@Google @GeminiApp @antigravity @GoogleAIStudio @GoogleDeepMind Looking forward to the new modelssss 👀

Ready, set, #GoogleIO. 🏁 Tune in tomorrow to hear our latest company-wide product updates and AI breakthroughs across Search, @GeminiApp, @Antigravity, @GoogleAIStudio, @GoogleDeepMind and more.

@danyurkin @speedrun @atomic_chat_hq You'll get it 🩷

just applied to @speedrun with @atomic_chat_hq! we're building local ai you can actually use - for humans and the agents they run. open-source, macos, windows, ios fingers crossed 🤞

@askOkara It broke when I tried to run it :(

this is one of the easiest ways to find backlink opportunities > go to okara.ai > drop your website url > ask it to find listicles and roundups in your niche where competitors are listed > reach out and ask to be included

@Sigma_Browser Which model would work best for sales-related data scraping?

Hermes agent running in private browser on free local models Private by design. Local. Open-source

Makes much more sense than ChatGPT's Atlas imo

Multi-Token Prediction (MTP) for LLaMA.cpp! Running Gemma4 local model 1.5x faster. We patched LLaMA.cpp. Quantized Gemma 4 assistant models into GGUF format. We ran tests on a MacBook Pro M5Max. Gemma 26B with MTP drafts tokens 40% faster. Benchmarks, source code and models 👇

zyphra just released open weights that beat models 15x their size on math benchmarks their new model is called zaya1-8b. mixture-of-experts, only 760m active params out of 8.4b total. apache 2.0, fully open. the benchmarks don't make sense for something this small: • aime '26: 89.1 – elite us math olympiad problems • hmmt feb '26: 71.6 – harvard-mit math tournament • imo-answerbench: 59.3 – international math olympiad level • livecodebench-v6: 65.8 – fresh coding problems, post-training-cutoff • gpqa-diamond: 71.0 – phd-level bio/chem/physics questions it's beating models with 10x more active parameters on hard math the scaling comparison is the real story zaya1-8b (0.7b parameters active) vs mistral-small-4 (119b parameters total): - aime: 89.1 vs 86.4 ✅ - hmmt: 71.6 vs 70.6 ✅ a model that fits on your phone winning against a 119b param model on competition math how? a few genuine technical innovations: • most models are pretrained on raw text, then reasoning is added in post-training via fine-tuning. zyphra included reasoning traces during pretraining itself – so the model learned to think step-by-step from the ground up, not as a patch on top • post-training runs four sequential rounds of trial-and-error learning. the model attempts problems, gets scored on how well it did, and updates itself based on that score – no human labeling needed. first round warms up basic reasoning, second throws increasingly hard problems at it, third focuses purely on math and code, fourth polishes how it behaves and follows instructions. each round specializes the previous one instead of trying to do everything at once • "markovian rsa" – when the model generates multiple candidate answers and picks the best one, each candidate carries a long chain of reasoning behind it. normally that eats through your context window fast. zyphra's fix: instead of keeping the full reasoning history for each candidate, only keep the last chunk. the model forgets the early scratch work but remembers where it ended up. same quality, fraction of the memory cost. the old playbook was simple: more parameters = smarter model. throw more compute, hire more engineers, train bigger zaya1-8b breaks that assumption. if a 65-person startup can match 119b-parameter models by being smarter about how they train, parameter count stops being a moat. architecture and training methodology are follow @thehypedotnews for daily analysis and breakdowns in ai

@GoogleDeepMind @EveOnline We'll see agents mining on orcas across the galaxy? Or stealing our assets on the way to jira?

We’re partnering with the developers of @EveOnline to explore the next frontier of AI research in games. EVE's complex, player-driven universe is the perfect safe sandbox to test agents on memory, continual learning, and long-term planning. Find out more → goo.gle/4epQIdy

@Sigma_Browser Awesome 👏🏻

Private AI browser with the OpenClaw agent on free local models Run your agent on Qwen, Gemma, or Nemotron directly in the browser Open source. Private. Runs on your local device

@Lovable I love lovable ❤️💙💜

Introducing the Lovable mobile app. Your ideas won’t wait. Now you can build them anywhere.

❤️💜💙

@quschdnjs @DeRonin_ It is never enough

@DeRonin_ 一天API够用？

One day with GPT 5.5 API better than 1 month without it Whenever with this tool you can take API for one day and don’t pay $100 per month We’re cooked

@BalaiBB @DeRonin_ I believe it’s great that we have both

@DeRonin_ claude or gpt which one is the goat ?

This is one way to connect your OpenClaw or Hermes agent to the new GPT-5.5 model (there are another 400 models available here, so take your pick). You can also go directly, but here you have a single API key to access every model.

@rugbist_ @svpino If you want - I can build that agent for you

@svpino yall been saying "faster than espresso" about every model launch since gpt-3. still no one shipped an agent that actually pays my rent. real question is how fast does it hallucinate my api key

As always - amidst the first

@AlexFromAtomic @atomic_chat_hq When we'll have it on board - I will :)

@schwarzerrttr @atomic_chat_hq build and share the results

Deepseek V4 Pro vs GPT-5.5 in a gamedev contest (full prompt is below)🏎️ Cost: Deepseek V4 Pro: $0.07656 GPT-5.5: $0.33063 Output stats: Deepseek: 34 tok/s · 9m 5s · 18,869 tokens GPT-5.5: 25 tok/s · 7m 5s · 10,580 tokens Conclusion: GPT-5.5 clearly made the better karting game. Deepseek V4 Pro was 4.3x cheaper and generated almost 2x more tokens, but the final result was weaker. It struggled with graphics, visual polish, and creative direction, while GPT-5.5 delivered better game quality, better visuals, more creativity, and stronger overall execution. Even though Deepseek positions itself as a strong model for coding, in this gamedev test it still felt far behind GPT-5.5. Try the same karting prompt with another AI model and share your result below.

1,556 commits in 7 days. one of the most important hermes releases yet - full react + ink rewrite of the terminal ui makes hermes feel more like a real product, not just a developer tool - native aws bedrock support makes enterprise deployment much easier, letting companies run hermes directly on top of Amazon Web Services infrastructure - expanded plugin hooks mean more custom tools, workflows, and integrations without changing the core system - orchestrator subagents let one main agent delegate work to smaller specialized agents instead of handling everything alone the scale is also very different: 1,556 commits, 1,320 files changed, and 96 contributors. v0.10.0 had 339 commits, 453 files changed, and 81 contributors that means +359% more commits, +191% more files changed, and +19% more contributors the most important change is the new transport architecture: providers are no longer tightly hardcoded into the agent loop. that makes model support faster, routing cleaner, enterprise deployment easier, and the product much harder to replace this is not just more features. it is a move from agent wrapper to agent platform

@d4n_nnyy @atomic_chat_hq I still love opus -_-

@atomic_chat_hq i've tried the same prompt with LAME opus 4.7. It is the middle between deepseek and gpt results. but the costs for opus 4.7 is same as gpt...