vibehacker
News
Google AI for Developers ·

Nano Banana 2.1 goes GA in the Gemini API at half Nano Banana 2's price, and the old model shuts Oct 29

Google's new image generation and editing model (gemini-nano-banana-2.1) shipped Oct 6 with better text rendering, up to 14 reference images, and Search grounding at $0.0336 per 1K image, $0.0504 at 2K, and $0.0756 at 4K, about half of Nano Banana 2, with batch halving that again. If your app calls gemini-3.1-flash-image, switch the model ID before it's shut down on Oct 29; there's no free API tier for 2.1.

Read at ai.google.dev
Google Gemini
Personal AI assistant from Google

More news

View all

Claude Code 2.1.292 lets you set a sub-agent's effort and switches stdio MCP servers to the 2026-07-28 protocol

The Oct 6 release adds an effort parameter to the Agent tool, a marketplace flag on claude plugin install , and fixes saved scheduled tasks never firing after /resume , /branch , or /clear , plus a security bug where hook approvals and auto mode skipped the prompt for file reads from network (UNC) paths. Local stdio MCP servers now negotiate protocol version 2026 07 28 by default, so if an older server breaks, set MCP PROTOCOL NEGOTIATION=legacy…

GitHub / anthropics/claude-code releases

Anthropic opens Claude cyber access to open-source maintainers and solo researchers in three new tiers

The expanded Cyber Verification Program, announced Oct 6, folds Project Glasswing in and offers Defense, Red Team, and Specialized tiers with fewer cyber blocks on Opus 5.5, Sonnet 5.5, and Mythos 5.1, and the Defense tier takes open source maintainers and individual researchers with a record of reported vulnerabilities, with replies in a few days. Enrolled orgs must accept data retention, while regular Claude models still handle code review, patching known issues, and finding bugs in code you own…

Anthropic News

Encrypted instructions on a web page can make Copilot CLI in autopilot leak your local secrets

Adversa AI says a page with ciphertext the agent decrypts in its own shell got Copilot CLI in autopilot to read files like .env.prod and send them to an attacker in 28 seconds, working on Microsoft's mai code 1.1 flash in half its runs while two GPT 5.6 models refused, and Auto routing can hand you the weak model without telling you. GitHub validated the report but says it isn't a vulnerability because the user asked for the fetch, so don't point an autopilot agent at untrusted pages from a directory that holds secrets…

Adversa AI Blog

GitHub stacked pull requests go GA, with `gh stack` now supporting Git worktrees

As of Oct 6, every github.com plan can split a big change into a stack of small PRs that are reviewed separately and land through the merge queue as one group, with Rebase stack now keeping approvals on unchanged code and auto merge for stacks rolling out over the next few weeks. The gh stack CLI extension now works with Git worktrees, which makes it easier to have parallel agents each push one layer of a stack…

GitHub Changelog

Spotted something we missed? Start a thread.