Gemini 3.5 Flash lands on OpenRouter, AllenAI ships OlmoEarth v1.1
Google's Gemini 3.5 Flash is now live on OpenRouter. AllenAI also dropped OlmoEarth v1.1, a more efficient open climate-science model family worth knowing about.
Google's Gemini 3.5 Flash is now live on OpenRouter. AllenAI also dropped OlmoEarth v1.1, a more efficient open climate-science model family worth knowing about.
Google's Gemini 3.5 Flash is now available on OpenRouter — meaning you can call it today via a single API endpoint without waiting for a separate Google AI Studio setup. That matters for indie builders who already route multiple models through OpenRouter and want to drop in a fast, cheap Google model without changing their stack.
Gemini 3.5 Flash is listed as a new model on OpenRouter under the handle google/gemini-3.5-flash. Flash models in the Gemini line are Google's speed-and-cost tier — designed for high-volume tasks like summarization, classification, and chat where you don't need the heaviest reasoning. If you're building a tool that calls an LLM on every user action, Flash is the tier to benchmark first.
AllenAI published OlmoEarth v1.1 on Hugging Face. It's a family of open models tuned for Earth and climate science tasks — think geospatial reasoning, climate data interpretation, and environmental question answering. Version 1.1 focuses on efficiency: smaller compute footprint for similar task performance compared to v1.0. The weights are open and downloadable now. If you're building anything in the climate-tech or environmental-data space, this is worth a look before defaulting to a general-purpose model.
The Model Context Protocol servers repo (modelcontextprotocol/servers) is sitting at 85,931 GitHub stars. That's not a new release, but it's a signal of how fast MCP adoption is moving. If you haven't looked at MCP yet, it's the protocol that lets AI agents talk to external tools and data sources in a standardized way — think of it as a USB standard for AI tool-calling. The repo contains reference server implementations you can run locally or deploy.
Continue (continuedev/continue) is an open-source AI coding assistant that runs inside VS Code and JetBrains. It's at 33,272 stars. It supports local models via Ollama as well as cloud APIs. Worth knowing if you want a Copilot-style experience without sending your code to a third-party server.
Three papers surfaced on Hugging Face today. SCICONVBENCH tests how well LLMs handle multi-turn clarification conversations in computational science — useful if you're building research assistants and want to know which models actually ask good follow-up questions. OSCAR introduces a technique for compressing the KV cache in transformers down to 2-bit precision without wrecking quality — relevant if you're running large models on limited hardware. DexHoldem is more of a curiosity: it's about embodied AI systems playing Texas Hold'em, which is a proxy task for physical dexterity and strategic reasoning.
1. Swap in Gemini 3.5 Flash on OpenRouter for any high-volume summarization or classification task you're currently running on a pricier model. Compare latency and cost over 100 calls — the Flash tier is built for exactly this.
2. If you work in climate tech or environmental data, download OlmoEarth v1.1 from Hugging Face and run it against your domain-specific questions. Compare it to a general-purpose model on the same prompts — you may get better answers at lower cost.
3. Browse the MCP servers repo and pick one reference server to run locally this weekend. Connect it to a simple agent using Continue or any OpenRouter-backed script. It's the fastest way to understand what the MCP hype is actually about.
Confirmed: Gemini 3.5 Flash is listed as a new model on OpenRouter at the URL google/gemini-3.5-flash as of today's signal data.
Not independently verified by CBW: We have not tested Gemini 3.5 Flash's actual latency, pricing tiers, or context window limits. OpenRouter's model card is the source — check it before building.
Confirmed: OlmoEarth v1.1 is published on Hugging Face by AllenAI with open weights, per the Hugging Face blog post.
Not independently verified by CBW: We have not benchmarked OlmoEarth v1.1 against v1.0 or against general-purpose models on climate tasks. AllenAI's efficiency claims come from their own blog post.
Worth noting: The MCP servers and Continue star counts are real GitHub numbers from today's signals, but star counts don't equal quality or active maintenance — check commit history before depending on either in production.
Source: OpenRouter — Gemini 3.5 Flash model page — https://openrouter.ai/google/gemini-3.5-flash
Source: Hugging Face blog — OlmoEarth v1.1 by AllenAI — https://huggingface.co/blog/allenai/olmoearth-v1-1
Source: Hugging Face Papers — SCICONVBENCH — https://huggingface.co/papers/2605.18630
Source: Hugging Face Papers — OSCAR 2-bit KV Cache — https://huggingface.co/papers/2605.17757
Source: GitHub — modelcontextprotocol/servers — https://github.com/modelcontextprotocol/servers
Source: GitHub — continuedev/continue — https://github.com/continuedev/continue
Get tomorrow's best AI project in your email. With a guide that works. Free. No spam.