Skip to main content
Back

Google Gemini 3.1 Pro: 2x reasoning boost, 77% ARC-AGI score

On saving token costs in Claude API, cross-platform sandbox in Cursor, Google's Pomelli generates images for marketing.

February 20, 2026 4 min read

Hey,

Your daily briefing is ready. You can finally take a break from the AI firehose.

Our algos spent the night splitting signal from noise and pulled the top news, models, papers, and repos.

Here’s the must-read:

Together with
Sponsor Logo
Summary

Read time: 4 min 37 sec

Top News

▸ Google rolls out Gemini 3.1 Pro with flexible reasoning modes

ElevenLabs

▸ Learn enterprise voice AI architectures and deploy production agents faster

Top News

▸ Anthropic introduces auto prompt caching for Claude API

Augment Code

▸ Build and coordinate AI agents faster using Augment Code’s unified workspace

Top Paper

▸ Cursor presents secure cross-platform sandboxing for local coding agents

Signals

▸ Google upgrades Pomelli with studio-style Photoshoot feature
▸ TinyFish scores 90% on Mind2Web, beating Gemini, OpenAI, and Anthropic
▸ Anthropic brings enterprise-secure Claude editing to Microsoft PowerPoint
▸ Replit lets users create animated videos by prompting in Animation mode
▸ OpenAI improves debugging with interactive code blocks in ChatGPT
Author photo
Today's Author
Lior Alexander. Founder of AlphaSignal and former ML Engineer. Previously built ML systems at Iguazio, Guesty, Enphase, Mila.
Top News
Google releases Gemini 3.1 Pro, doubling ARC-AGI-2 reasoning performance over 3 Pro
48,295 Likes
Grok 4 Fast Benchmark

Google releases Gemini 3.1 Pro as the upgraded core model behind the recent Gemini 3 Deep Think update. It scores 77.1% on ARC-AGI-2, more than double Gemini 3 Pro’s 31.1%. It now rolls out across consumer apps, developer tools, and enterprise platforms.

Core Reasoning Upgrade

Gemini 3.1 Pro improves baseline reasoning for complex problem-solving. On ARC-AGI-2, which tests a model’s ability to solve unfamiliar logic patterns, it reaches 77.1%. That places it ahead of prior Gemini versions and several competing frontier models on this benchmark.

Google keeps API pricing unchanged compared to 3 Pro and maintains a 1M token context window.

Designed for Complex Tasks

3.1 Pro targets problems that require structured reasoning rather than short answers. It supports:

  • Advanced logical problem solving

  • Scientific and technical reasoning

  • Competitive coding tasks

  • Agentic search workflows

  • MCP tool usage

It also generates website-ready animated SVGs directly from text prompts. These outputs use pure code, remain scalable at any resolution, and keep file sizes small.

Availability

Google ships 3.1 Pro across its ecosystem:

  • Developers: Gemini API in Google AI Studio, Gemini CLI, Google Antigravity, Android Studio

  • Enterprises: Vertex AI and Gemini Enterprise

  • Consumers: Gemini app and NotebookLM (Pro and Ultra plans with higher limits)

Build with 3.1
Presented by ElevenLabs
Voice AI Is Moving From Experimentation to Enterprise Deployment
alpha_signal_image_1

In this new handbook, ElevenLabs breaks down

  • The Architectures

  • Trade-Offs

  • Real-World Case Studies

Behind large-scale conversational agent implementations, helping leaders choose the right path to quality, compliance, and speed.

Download Building Voice Conversational Agents at Scale and see how leading enterprises are making AI conversations real.

Download handbook
partner with us
Top News
Anthropic enables auto prompt caching in Claude API to cut repeated token costs by 90%
5,829 Likes
Grok 4 Fast Benchmark

Every AI agent repeats itself. It resends system prompts, tool definitions, and conversation history on every turn because the Claude Messages API is stateless.

Anthropic now removes the cost of that repetition with auto prompt caching, cutting cached input token costs to 10% and reducing latency on repeated context.

Here is the core idea. Large language models process prompts in two steps: prefill (read the prompt) and decode (generate output).

Caching stores the prefill computation for identical prompt prefixes. If the same content appears again, Claude reuses the stored computation instead of recalculating it.

Key features and results:

  • Cached input tokens cost 10% of normal input tokens

  • Reduces latency on repeated prompt sections

  • Works automatically with one cache_control parameter

  • Moves the cache breakpoint as conversations grow

  • Matches exact content using cryptographic hashing

How to use it:

  • Add "cache_control": to your API request

  • Place static content (system prompts, tools) first

  • Append new messages instead of editing history

  • Avoid small text changes that break cache matches

Turn on caching
Presented by Augment Code
Intent: The workspace built for agent orchestration
alpha_signal_image_2

Intent is a developer workspace for engineers shipping with AI agents.

Each task gets a workspace with a living spec, agent notes, and transparent file changes, so you can coordinate specialists, avoid conflicts, and move fast without juggling panes and branches.

Build with intent
partner with us
Top Paper
Cursor deploys a cross-platform sandbox that reduces agent interruptions by 40% while enforcing strict execution boundaries
2,826 Likes
Grok 4 Fast Benchmark

Coding agents now run terminal commands, edit files, and execute scripts. When you auto-approve those commands, you gain speed but accept risk. An agent can delete files, modify git configs, or expose secrets. Cursor addresses this with a local sandbox across macOS, Linux, and Windows (via WSL2).

The sandbox lets agents operate freely inside a controlled environment and only request approval when they need network or elevated access. Sandboxed agents stop 40% less often, which reduces approval fatigue.

The sandbox uses OS-level controls such as Seatbelt (macOS), seccomp (blocks unsafe system calls), and Landlock (restricts file access).

Key features and results

  • Fine-grained filesystem restrictions, including .cursorignore

  • Blocked unsafe syscalls and sensitive config writes

  • Explicit permission escalation when required

  • 40% fewer interruptions versus unsandboxed agents

  • Cross-platform support

How to use it

  • Enable sandboxing in supported environments

  • Let agents run normally inside the workspace

  • Approve only when they request external access

Enable sandbox
Signals
1 Google Labs unveils Photoshoot, a new feature in Pomelli for AI-powered product marketing images 34,857 Likes
2 TinyFish topped the Online Mind2web bench at 90%, outperforming Gemini by 21 and OpenAI by 29. All runs made 100% public. Presented by TinyFish
3 Anthropic introduces Claude in PowerPoint with template-aware slide generation 8,224 Likes
4 Replit enables AI-generated animated videos directly inside its App tab powered by Gemini 3.1 Pro 3,295 Likes
5 OpenAI upgrades Code Blocks with live previews, split-screen editing, and full-screen mode 1,629 Likes
At Alpha Signal, our mission is to build a sharp, engaged community focused on AI, machine learning, and cutting-edge language models, helping over 200,000 developers stay informed and ahead. We’re passionate about curating the best in AI, from top research and trending technical blogs to expert insights and tailored job opportunities. We keep you connected to the breakthroughs and discussions that matter, so you can stay in the loop without endless searching. We also work closely with partners who value the future of AI, including employers and advertisers who want to reach an audience as passionate about AI as we are.

Our partnerships are based on shared values of ethics, responsibility, and a commitment to building a better world through technology.Privacy is a priority at Alpha Signal. Our Privacy Policy clearly explains how we collect, store, and use your personal and non-personal information. By using our website, you accept these terms, which you can review on our website. This policy applies across all Alpha Signal pages, outlining your rights and how to contact us if you want to adjust the use of your information. We’re based in the United States. By using our site, you agree to be governed by U.S. laws.
Looking to promote your company, product, service, or event to 250,000+ AI developers?
Work with us