Google Launches Gemini 3.6 Flash and 3.5 Flash-Lite for Production Use

·1 views

Google has made Gemini 3.6 Flash and Gemini 3.5 Flash-Lite generally available for production use. Gemini 3.6 Flash is priced at $1.50 per million input tokens and $7.50 per million output tokens, offering stronger performance on complex agentic and multimodal tasks with reduced token usage compared to its predecessor. Gemini 3.5 Flash-Lite, the faster and cheaper option at $0.30 per million input tokens, is designed for high-throughput workloads and outperforms prior Flash-Lite generations on reasoning and multimodal benchmarks. Both models support a one-million token context window, 64,000 maximum output tokens, and built-in tools including Computer Use. Developers can migrate existing applications to either model using the updated Gemini Interactions API.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Why NVMe Storage Is Making Hardware RAID Controllers Obsolete in Data Centers

For years, hardware RAID controllers were the standard for disk protection in enterprise environments, but NVMe technology has fundamentally changed that calculus. NVMe drives communicate directly over the PCIe bus, and routing multiple high-speed drives through a single hardware RAID controller's PCIe slot can cut aggregate throughput by up to 50%. Modern software RAID solutions like Linux mdadm or ZFS allow NVMe drives to connect directly to the motherboard, leveraging the hundreds of PCIe lanes available on AMD EPYC and Intel Xeon processors. Enterprise-grade NVMe drives also include built-in Power Loss Protection capacitors, negating the data-safety argument traditionally used to justify hardware RAID controllers. Experts now recommend software RAID for NVMe deployments, citing better performance, no licensing costs, and freedom from vendor lock-in during disaster recovery.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Developer Builds 178-Line Python Firewall to Block Scam Calls at Carrier Level

A developer has published an open-source call-screening system called the Edge Fraud Firewall, designed to intercept fraudulent calls before they reach an application or human agent. Built in 178 lines of Python using a single Flask file, the tool integrates three Telnyx APIs — Number Lookup, AI Inference, and Call Control — to evaluate every inbound call at the carrier edge. Each call is assessed through a pipeline that checks a blocklist, looks up carrier data, and uses an AI model to classify the caller as CLEAN, SUSPICIOUS, or BLOCK. Legitimate calls are forwarded normally, blocked numbers are rejected and added to a blocklist, while suspicious callers are routed to a honeypot that loops hold music indefinitely to waste scammers' time and resources. The system defaults to CLEAN if the AI service is unavailable, ensuring legitimate callers are never incorrectly blocked due to a technical failure.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

How to Build Your First CI Pipeline Using GitHub Actions in 10 Minutes

GitHub Actions provides a free, built-in CI pipeline tool available to every GitHub repository, requiring no additional setup or cost. Developers can create a working pipeline by adding a YAML configuration file at .github/workflows/ci.yml, which triggers automated checks on every push or pull request. The pipeline can be configured to install dependencies using npm ci, run linting, and execute tests, displaying a green checkmark or red X directly in the GitHub interface. Key concepts include workflows, events, jobs, steps, and runners, which nest together to form the complete automation model. The same workflow structure applies across languages including Python and Go, with only the install and test commands needing adjustment.

0 comments Read more at DEV Community

ProgrammingDEV Community ·

Rule of Three: How indie developers can stop over-building and start shipping

Many indie developers stall their side projects by over-engineering features before validating the core product, a pattern one developer described after spending six months on a project that launched to zero user interest. The author argues that solo builders often treat personal projects like enterprise software, adding polish, scale, and edge-case handling before a single user has tested the app. To counter this, they propose the 'Rule of Three': identifying only the three core actions a user needs to solve the primary problem, and shipping nothing beyond that until the core loop is proven. Clean code and complex architecture are framed as secondary concerns, with messy but functional code preferred if it accelerates real-world validation. The recommended cycle is to build the minimal core, observe five real users interacting with it without assistance, and simplify based on what is actually confusing rather than layering on tooltips or tutorials.

0 comments Read more at DEV Community