AI

Google’s Gemma 4 family covers the 2.3B E2B, 4.5B E4B, 26B MoE, and 31B dense variants. It delivers strong open-weight performance across text, vision, and audio. But general-purpose models still struggle with narrow tasks. You often need a fixed output format, special terms, or facts that weren’t in the training data. Fine-tuning fixes this. Unsloth makes it work on a single consumer GPU. Its custom CUDA kernels cut VRAM by up to 60% and double training speed next to a standard Hugging Face plus PEFT setup. The same Unsloth path fine-tunes Chinese-lab coding models like Qwen and GLM.

For most developers in 2026, Gemma 4 31B is the best all-around open model. It ranks #3 on the LMArena leaderboard, scores 85.2% on MMLU Pro, and ships under Apache 2.0 with zero usage limits. Qwen 3.5 27B edges it on coding, and its Omni variant offers real-time speech output that no other open model matches. Llama 4 Maverick (400B MoE) wins on raw scale, but it needs datacenter hardware and Meta’s restrictive 700M MAU license. So pick Gemma 4 for the best quality-to-size ratio, Qwen 3.5 for coding-heavy work, and Llama 4 only when you need the largest open model. For pictures rather than text, Alibaba’s Qwen-Image tops the local image field .

You can build a fully local meeting transcriber on Linux. Capture system audio with PipeWire. Transcribe with Faster-Whisper on your GPU. Pipe the transcript to a local LLM through Ollama for structured summaries with names, decisions, and action items. The pipeline runs on 16GB of RAM and a mid-range NVIDIA GPU, and produces notes within seconds of the call ending. No data leaves your network.

Commercial services like Otter.ai and Fireflies.ai route your audio through their servers. If your meetings cover sensitive topics like product plans, HR, or legal reviews, that’s a non-starter. A local pipeline gives you the same structured output, and nothing leaves your building.

You can unify access to Ollama, vLLM, cloud providers like OpenAI, Anthropic, and Google, plus custom model servers behind one OpenAI-compatible endpoint using LiteLLM Proxy . LiteLLM is a reverse proxy. It maps the standard /v1/chat/completions request to each provider’s native API. From one YAML file it handles auth, model routing, load balancing, fallbacks, rate limits, and spend tracking. Your app calls one endpoint with one key, and LiteLLM picks the right backend. You can swap models, add providers, or run A/B tests without touching app code.

Three focused AI coding agents beat one broad agent working three times as long. Addy Osmani showed this at O’Reilly AI CodeCon , and the finding captures both the upside and the catch of multi-agent work. The speed gains are real. They only show up when you solve the coordination problem. Without file isolation, iteration caps, and review gates, parallel agents make a mess of merge conflicts and duplicated work.

In practice, the tooling breaks into three tiers. In-process subagents handle focused delegation in a single terminal. Local orchestrators run 3-10 agents with dashboard control. Cloud-async tools handle unattended overnight runs. Most developers use all three tiers daily, switching based on task size and whether they plan to stay at the keyboard.

Claude Code, Cursor, and GitHub Copilot take three very different shots at AI-assisted coding: a terminal-native agent, an AI-first IDE, and a multi-IDE plugin. Claude Code leads on raw skill and complex multi-file work, scoring highest on SWE-bench at about 74-81%. Cursor offers the best editor experience with background agents and cloud automation. GitHub Copilot has the lowest entry price at $10/month and the widest IDE support. Most pro developers now mix two or more tools, with Claude Code plus Cursor as the top pair per the JetBrains AI Pulse survey from January 2026.