Gemini 3.1

1. Qwen3.6 35B: paratroopers puzzle (www.reddit.com)

   +51 9w gemini

   I keep presenting Local and Huge cloud models with the same challenge: "Two paratroopers land on an infinite 1D numeric axis at distinct, unknown integer coordinates. They both execute the exact same deterministic program.

2. FrontierMath: Opus 4.7 improves over Opus 4.6 and Gemini 3.1 but still trails GPT-5.4-xHigh and GPT-5.4-Pro (www.reddit.com)

   +406 9w gpt-5geminiopus

   could not extract summary

3. SFT + DPO on open-sourced SLMs (www.reddit.com)

   +75 9w dpogpt-5deepseek+2

   Hey folks, this is for those who appreciate experimentation on open-sourced AI models. We fine-tuned open-sourced SMLs (3B and 7B parameters) with SFT + DPO against commercial models like GPT-5.4, Gemini 3.1 Pro, Claude Opus 4.6, Google Do…

4. New “pelican test” but for video (www.reddit.com)

   +1 10w gemmaqwengemini

   If the LLM supports video—which most VLLMs nowadays do—then try the following prompt with the accompanying video: With the given video, which is about 16 seconds long, your task is to write JavaScript for an animation that faithfully repli…

5. Show HN: Claude Opus 4.7: Everything You Need to Know (news.ycombinator.com)

   +11 10w tool-usegpt-5mythos+4

   Claude Opus 4.7 is Anthropic's most capable generally available model, released April 16, 2026. It outperforms Opus 4.6, GPT-5.4, and Gemini 3.1 Pro on key benchmarks including agentic coding, multidisciplinary reasoning, scaled tool use,…

6. Gemini 3.1 Pro #1 at METR Timeline 80% Success Rate (1.5H) (www.reddit.com)

   +9122 10w gemini

   #2 at 50% success rate (task length: 6H 24M)

7. Gemini 3.1 Flash TTS – with directed prompts (simonwillison.net via hn)

   +62 10w gemini

   Google released Gemini 3.1 Flash TTS today, a new text-to-speech model that can be directed using prompts. It's presented via the standard Gemini API using gemini-3.1-flash-tts-preview as the model ID, …

8. Google Launches Gemini 3.1 Flash TTS Text-to-Speech Model (x.com via reddit)

   +319 10w gemini

   Logan Kilpatrick on X: "Introducing Gemini 3.1 Flash TTS 🗣️, our latest text to speech model with scene direction, speaker level specificity, audio tags, more natural + expressive voices, and support for 70 different languages. Available v…

   • Gemini 3.1 Flash TTS: the next generation of expressive AI speech (blog.google)
9. Show HN: Hormuz Trail - Oregon Trail parody/black-box AI coding exercise (hormuztrail.com via hn)

   +3 10w sonnetgeminicursor+2

   I jokingly told a co-worker Iran might make a good Oregon Trail parody. Then I built it.

10. Gemma 4 31B passed 7/8 real-world production tests — including ones I designed to make it fail. Full prompts + outputs. (www.reddit.com)

    +1211 10w moegemmagemini+1

    I've been waiting for a capable free local LLM for a while. I think we're close — the quality is getting there fast, and Gemma 4 is the first open-weight model where I genuinely considered using it in production for simple-to-medium tasks.

11. Compare harnesses not models: Blitzy vs. GPT-5.4 on SWE-Bench Pro (quesma.com via hn)

    +1 10w swe-benchgpt-5gemini+2

    An independent audit of agentic scaffolding and harnesses. We analyze how agent workflows, codebase documentation, and test verification impact performance compared to raw base models like GPT-5.4, Gemini 3.1 Pro, and Claude Code.

12. Open-sourcing Dograh - our voice AI agent platform built as an alternative to Vapi (www.reddit.com)

    +1 10w gemini

    We are open-sourcing the backbone of our voice AI stack - Dograh, a self-hostable, open-source voice agent platform. Three core things that make it work: Visual Workflow Builder What it is: Drag-and-drop builder for designing voice agent c…

← all threads