Gemini

Gemini AI

Gemini AI

Gemini AI

Gemini AI

Gemini AI

Gemini AI

Gemini AI

Gemini AI

Gemini AI

Gemini AI

I made Claude Code pull my team into its planning loop (open source MCP server)

Anyone else notice that in planning mode, Claude Code constantly hits design forks — "queue or cron?", "which auth flow?", "REST or events?" — As a solo dev I'd either rubber-stamp it or jump into Slack to ask people, which kills the whole flow. So I built **shared-brainstorm**, an MCP server that brings teammates into the planning loop: - Claude Code hits a design question and routes it to a shared web page. - Teammates open a link and discuss right there — **no install, no signup, no account.** Just a link. - Claude reads the team's input and folds it into the plan, while you drive the whole thing from your terminal. The zero-install part is the point: your teammates never touch npm, never log into anything, never leave their tab. You run it locally — it spins up a local server + tunnel, so there's no SaaS and nothing to host. Free + open source, on npm as `shared-brainstorm`. Also works with Codex, OpenCode, and Gemini CLI. 60-sec demo: https://youtu.be/cP9V4pDTtVQ Repo: https://github.com/mohitmayank/shared-brainstorm _ Would love feedback from people who pair Claude Code with a team.

The credits run out quickly

Hello everyone. I have zero programming knowledge but seeing the boom that everyone was talking about Claude I started tinkering with it. I've used other IASs like Gemini, Deepseek, Chatgpt... but none as good in code as Claude. I feel like he never lies to me. He suggests really good ideas, and he always does what he says until it works, even if it's just a matter of reviewing static HTML for GitHub Pages. But I run out of credits quickly. I use the free version and all I'm asking is for him to review an index page he created for me (a simple website, nothing special), but since I don't understand how I sent it to him or how he modified the index page , it makes me wait 5 hours.. I always work on the same conversation (maybe that's another problem) so I'm asking for advice in case I ever pay for the pro plan, so I don't waste it two days later. Would using Claude Code help at all instead of the web version?

ChatGPT-5.5 Beats Opus in Realistic Benchmark (DeepSWE)

From the website, it touts: * Contamination free: Tasks are written from scratch, not adapted from existing commits or PRs, so no model has seen the solution during pretraining. * High diversity: Tasks span a broad pool of 91 repositories across 5 languages. * Real-world complexity: Prompts are ~half the length of SWE-bench Pro's, yet solutions require 5.5x more code and ~2x more output tokens. * Reliable verification: Verifiers are hand-written to test software behavior rather than implementation details. And the scores match more with actual experiences when using an LLM to do real coding. For example, Gemini 3.1 Pro tends to score decently on SWEbench Pro although we all know it can't do a thing. On this benchmark, it scored ~18%. Mythos needs to come out! It seems that ChatGPT-5.5 is the current king of real code changes. Opus lags a bit... 70% for GPT versus 54% for Opus. There is a lot of criticism of SWEbench Pro and the scores on it discussed in fine detail. A lot of interesting stuff. For example, SWEbench Pro prompts tell the LLM not to write tests. Claude goes ahead and writes them ~20% of the time whereas GPT only did it ~10% of the time. By not following instructions, Opus could pull ahead in some of the test cases in that way. In deepSWE, the test prompts don't specify, so you see more what the LLM chooses to do when given a challenge. Both GPT and Opus went ahead and wrote tests 80-90% of the time, a good thing for it to do in general. I can't overstate the correction here telling the whole story if you don't want to read deeply into the methodology and critiques of SWEbench Pro. If you want a tl;dr, look at the graph of [results here](https://deepswe.datacurve.ai/blog#results). On the left, you have scores on SWEbench Pro, and on the right, you have scores on deepSWE. We see a large correction in the direction that matches our real experiences when using LLMs to solve actual multi-step coding problems. I mean, Haiku at 30%? Nah, it's more like 0% as it should be. I already mentioned Gemini 3.1 Pro dropping from competitive to absolute garbage, and that matches how no programmer uses anything other than Codex and Claude Code to do real work. GPt-5.4 and GPT-5.5 scoring about the same 58.5% on SWEbench Pro also makes no sense, but on this deepSWE, GPT-5.5 crushes GPT-5.4 going from 56% to 70%. The small models like Gemini 3 flash and Haiku-4.5 scoring up there at around 35-40%? More like 0% like it actually is. And this bench finally shows how much better Opus-4.7 is compared to Sonnet-4.6. Sonnet is still a great workhorse for simpler issues, but when it comes to the multi-step challenges in real codebases found in deepSWE, Opus gets a 54% versus Sonnet's 32%. Kimi 2.6, mimo v2.5 Pro, glm-5.1, and deepseek v4 pro all scored less than gpt-5.4-mini. Ouch. Open-weight models just can't code that well. One variable might be the prompting style in deepSWE versus SWEbench Pro. DeepSWE was much more natural. "Here's the issue, and I want it to do this." SWEbench Pro gave a prompt with like 10 steps in it, telling the model more so how it might want to approach a code change. Step 1, step 2, etc. Opus 4.7 scored 54% compared to 28% by Opus 4.6, so 4.7 was an actual large leep when it comes to barebone prompts in multifile, multi-step code changes. __Anthropic gang *needs* 2 CCs of Mythos STAT!__ PS Make sure you read the limitations section. There is no benchmark that is 100% perfect.

is claude pro worth it for a marketer?

I work in marketing and do a little bit of vibe coding. I currently use Gemini as my main LLM and I'm thinking about switching to Claude. Are the token limits in the $20 plan enough for my use?

Gemini explain please...

[https://gemini.google.com/share/1b2ff803d882](https://gemini.google.com/share/1b2ff803d882) I'm sorry earlier today i made a [post comparing ChatGPT and Gemini](https://www.reddit.com/r/artificial/comments/1tp7v4b/paid_gemini_vs_free_chatgpt/). I asked Gemini to build a prompt and gave it to him in another chat and i got this...

Which provider fits best for my needs?

Hi everyone, I’m looking to get more into experimenting with AI and considering a paid subscription, but I’m a bit unsure which direction makes the most sense for my use case. My main goals: \-Writing a technical book in the field of taxation \-Preparing presentations and structured content \-Learning and experimenting with programming \-Building automation workflows (e.g. n8n) \-Running or experimenting with tools like Hermes / OpenClaw (I know Claude doesn’t work everywhere there) \-Testing new AI features (e.g. Claude artifacts, coding tools, agents, etc.) From what I’ve read recently, opinions are all over the place: Some say ChatGPT (with Codex-style tools) is strongest for coding + general use Others argue Claude is better for writing and reasoning-heavy tasks Gemini seems strong for long context and Google integration And then there’s the API route (DeepSeek looks extremely cheap right now and seems attractive for experimentation) So I’m trying to figure out what actually makes sense in practice. Would you recommend: A ChatGPT subscription Claude Pro Gemini Advanced Or skipping subscriptions and going API-first with models like DeepSeek / others? Would really appreciate real-world experiences—especially from people doing a mix of writing + coding + automation rather than just one narrow use case. Thanks! (Ai generated as englisch is not my mother language)

Share you experience building a saas using ai

I tried about 5 times and each time i fail. It has been more than year trying and i'm getting frustrated. Here is my attempts: 1- Lovable: Insane UI, Bad functionality, Unable to migrate easily 2- Claude code + bmad method: Insane planning, Endless implementation with no real result. 3- Claude code + superpowers: It can't build a full app at all. but it perfect for single specific feature. 4- Claude code + GSD: This time i really got great output with very good tracking. the problem that i realized later is that the infrastructure is dump. 5- Pure claude code/opencode/gemini cli; Not usable at all. it is actually better at ui. but that's only that (usually) Time and UI represent the biggest obstacles for me. Please help me by sharing your advice or experience. Edit: I'm depending on Chinese models. can this be an obstruct?

PAID Gemini vs FREE ChatGPT

I recently subscribed to Google One Ai Pro and recieved Gemini Plus Plan... I've been using it for some days, and the difference between Gemini and ChatGPT is enormous... i feel like talking to an Ai model from 2022. I asked them both to generate an image using the EXACT same prompt, here are the results... The prompt: "Generate a creepy midnight image in an abandoned road and there is a scary woman with white - blue gown standing next to the road. Make the quality unremarkably iPhone-ish, slight motion blur, grainy quality as if it was taken in dark. The picture is taken from a car in motion, from it's window on the front right seat." Models used: Gemini 3.1 Pro GPT-4o (afaik this is the model used in image gen in the free ChatGPT version atm) Edit: Added the models used. https://preview.redd.it/bbou5fdr0p3h1.png?width=1340&format=png&auto=webp&s=46ff98af1e386f8a4da6b1c304e13d1b319b95e5

Help getting a workflow to work properly

Coming out of a long day of back-to-back meetings, I had an idea to use Claude to help me keep track of things. The general idea is that I could write a skill that I would invoke "/evening-ritual" and Claude would peruse through my Gmail and Calendar, looking at all of the meetings I sat in and the emails that I sent/received. We use Gemini Notes/Transcripts for \*most\* of our meetings at work, so it would match those up. Then, I could hit "Voice Mode" and have it talk through my day with me, going meeting by meeting. For the ones where it has a transcript, we would talk through any action items or things I need/want to remember. For meetings without a transcript, it would ask me for things I remembered or might've written down physically, etc. It would then produce an overview of my day - key decision points, any open loops, things I need to come back to, action items, etc, and drop it into a markdown file that would get created/pushed to my Obsidian Vault. The idea is that then, I could have a similar morning routine that would recap things that are pressing from the previous day, or upcoming important meetings I should prep for (anything with less than 4 people OR a meeting with an attendee outside our company). This seems easy enough, but doing it via Claude Chat was an exercise in frustration: * It had A LOT of trouble finding transcripts; notably, ones that I had already marked as "read" in my inbox. It also seemed to not understand that "Gemini Meeting Notes" included notes \*and\* a transcript * It skipped meetings, and I had to remind it to go chronologically through the day * Even when I gave it the transcript directly, it seemed to struggle to find action items *for me*, and twice it asked me to summarize the meeting instead of it reading the transcript I had just provided, "to ensure it didn't misread anything". It was also frustrating trying to use voice mode but then also sometimes trying to give it a link to a document and then enter back into voice mode. Anyone got any ideas to better solve for this? I know I could build something like this in n8n, but I really didn't want to spin all that up when this seemed like such an easy Claude task. Should I try it in Cowork instead of Claude Chat?

https://t.co/j6Qu6uAg1b

https://t.co/j6Qu6uAg1b

Built a tool to save Claude responses (and ChatGPT, Gemini) into one searchable vault - sharing in case it's useful

I built this tool because I kept asking Claude for code and explanations and losing them in long chats. Coffer adds a save button to every AI response and stores them locally in a searchable vault. Works on: \- [claude.ai](http://claude.ai) \- [chatgpt.com](http://chatgpt.com) \- [gemini.google.com](http://gemini.google.com) You can mix snippets across all three and search them. The Markdown stays formatted, which is very nice for Claude's longer responses with code and tables. Everything is local. Coffer makes zero network calls of its own. Free. Feedback is especially welcome. [https://chromewebstore.google.com/detail/nhchbmaobjhjfmeekpnkmhdjajdolcjb?utm\_source=item-share-cb](https://chromewebstore.google.com/detail/nhchbmaobjhjfmeekpnkmhdjajdolcjb?utm_source=item-share-cb)

OpenAI image generation is just superior to any other tools and it's not even close

I've been running a little experiment where I ask Chatgpt and Gemini to generate the same image for about a month, and not a single time I got a better result from Gemini. I have a pro account with both and I see people giving so much praise to Nanobanana and it makes me wonder where it's coming from

The Singularity Gate – a new benchmark for AI predicting post-cutoff scientific discoveries

I just released a new benchmark called The Singularity Gate. Tests whether frontier AI can predict paradigm-breaking scientific discoveries published after their training cutoff. **Top score:** 17.75% (partial credit, Opus 4.7). **Fully-correct outcome rate:** 0% across all respondents. Passing the Singularity Gate is necessary, though not sufficient, for autonomous AI-driven discovery. A model that can predict paradigm-breaking discoveries isn't necessarily Einstein-level. But a model that can't is definitely not. 1. Claude Opus 4.7 (max) - 17.75% 2. GPT-5.5 (xhigh) - 16.08% 3. Claude Opus 4.6 (max) - 15.11% 4. Gemini 3.1 Pro (high) - 14.42% 5. Claude Sonnet 4.6 (max) - 13.67% These are partial-credit scores. No model fully predicts a discovery. Happy to discuss methodology, related work, or the framing in the comments. **Paper:** [https://doi.org/10.5281/zenodo.20358378](https://doi.org/10.5281/zenodo.20358378) **Website:** [https://singularitygate.org](https://singularitygate.org)

AI quality/usage over 90 min chat, mostly Q&A, summaries and conclusions.

I compared ChatGPT (Plus - Auto), Claude (Pro - Sonnet 4.6) and Gemini (Pro - Flash) over 90 minutes, mostly Q&A about mobile phones, asked to research specs, reviews, pros and cons, create executive summaries with the results, etc., nothing complex, I stayed in the same conversation/context the whole time. At 90min, Claude 0% left, ChatGPT 99% and Gemini 100%. I have to say the quality and design/formatting of the Claude output is amazing, the results/conclusions the same across the board.

The Singularity Gate – New Benchmark for AI predicting post-cutoff scientific discoveries. Opus 4.7 is in the Lead

I just released a new benchmark called The Singularity Gate. Tests whether frontier AI can predict paradigm-breaking scientific discoveries published after their training cutoff. **Top score:** 17.75% (partial credit, Opus 4.7). **Fully-correct outcome rate:** 0% across all respondents. Passing the Singularity Gate is necessary, though not sufficient, for autonomous AI-driven discovery. A model that can predict paradigm-breaking discoveries isn't necessarily Einstein-level. But a model that can't is definitely not. https://preview.redd.it/lywtnl5zbh3h1.png?width=900&format=png&auto=webp&s=c3211eddfb5fcaaf60bb549e5ce0e66770db14ed 1. Claude Opus 4.7 (max) - 17.75% 2. GPT-5.5 (xhigh) - 16.08% 3. Claude Opus 4.6 (max) - 15.11% 4. Gemini 3.1 Pro (high) - 14.42% 5. Claude Sonnet 4.6 (max) - 13.67% These are partial-credit scores. No model fully predicts a discovery. Happy to discuss methodology, related work, or the framing in the comments. **Paper:** [https://doi.org/10.5281/zenodo.20358378](https://doi.org/10.5281/zenodo.20358378) **Website:** [https://singularitygate.org](https://singularitygate.org)