AI news for Wednesday, October 7, 2026: Google ships Nano Banana 2.1, limits free Gemini to Flash-Lite

Google gives with one hand and takes with the other, OpenAI makes decisions cheap, and the real question everywhere is who checks the output.

Creative Models3 min
Nano Banana 2.1 holds characters and infographics together
Google released Nano Banana 2.1 on October 6, an update to its Flash-tier image model that, by Google's own evaluations, holds multi-character scenes together far better. If you make images with people in them or infographics with real text, test it this week.
Why it matters If you generate branded visuals, comics or infographics, the weak spots were consistent characters and readable layouts. Google says both improved a lot. Rerun your hardest prompts before you trust the numbers.
SRCGoogle AI for Developers
Products3 min
Google limits free Gemini to Flash-Lite from October 9
From October 9, the free Gemini app offers only Flash-Lite, and Google AI Plus subscribers lose Pro on a date Google will email them, per Google's help center. Free Gemini is now a sample. Find out this week whether the sample does your work.
Why it matters If free Gemini does real work for you, Friday is the day it gets smaller. Run your usual tasks on Flash-Lite before then and decide with evidence whether you need Flash (AI Plus) or Pro (AI Pro).
SRCGoogle
Products3 min
OpenAI opens its Decisions API, and output is free
OpenAI put its Decisions API in public beta on October 6, with gpt-6-luna at $0.10 per million input tokens and no charge for output. If you classify, route or grade at volume through a chat model, this is the endpoint to price against.
Why it matters Moderation, ticket routing and rubric grading are mostly yes, no or pick-one questions, and many teams answer them with a full chat model. Run a day of that traffic through the Decisions API and compare the bill and the latency before the beta turns GA.
SRCOpenAI
Safety Policy3 min
Anthropic splits Claude cyber access into three tiers
Anthropic expanded its Cyber Verification Program on October 6 into Defense, Red Team and Specialized tiers that lift some of Claude's cyber restrictions for vetted teams. If refusals have slowed your security work, verification is now the way around them.
Why it matters If you run a SOC, reverse malware or maintain open-source code, Defense Access is the tier to apply for, and Anthropic says answers take a few days. Pen testers need an organization behind them and should expect a review of a few weeks.
SRCAnthropic
Research Models3 min
OpenAI posts 722 math papers from an unreleased model
OpenAI published 722 manuscripts in 372 result families on GitHub on Tuesday, produced mostly by an unreleased internal model, and not all are formally verified. The bottleneck in AI math has moved from finding proofs to checking them.
Why it matters Machine-generated work now arrives faster than people can check it, in math first and in your codebase next. If you ship AI output, pair the generator with a checker, the way OpenAI paired some of these proofs with Lean.
SRCOpenAI (GitHub)

ShareX LinkedIn Hacker News Email

Prompt of the day

Find out which AI model tier your work actually needs

› I'm deciding which AI plan I actually need. Here are the tasks I used an AI chat assistant for in the last two weeks, one per line, with roughly how often:

[paste your list]

Ask me up to three questions first if the list is too vague. Then, for each task:
1. Rate how demanding it is for a model: light (rewriting, summarizing, short answers), medium (multi-step drafting, analyzing a long document) or heavy (complex code, deep research, math). Give the reason in one line.
2. Say whether a small, fast model would likely handle it well enough, and describe what a failure would look like so I can spot it.
3. Write one short test prompt I can run on a small model and on a larger one to compare the results side by side.

Finish with a short summary: how many of my tasks need a larger model, how often I hit them, and whether paying for a higher tier is worth it for me. Be blunt.

Open in Magai

Works in any AI chat. In Magai, run it against several models at once.

Get every edition

This one arrived in inboxes at 6am ET.

Nano Banana 2.1 holds characters and infographics together

Google limits free Gemini to Flash-Lite from October 9

OpenAI opens its Decisions API, and output is free

Anthropic splits Claude cyber access into three tiers

OpenAI posts 722 math papers from an unreleased model