← Reports

AI Radar

Thursday, May 21, 2026 · auto-generated by tp3_ai_radar.py

Top 3 today

How fast is 10 tokens per second really?

9/10

Simon Willison local LLM · May 20 · simonwillison.net

How fast is 10 tokens per second really? Neat little HTML app by Mike Veerman ( source code here ) which simulates LLM token output speeds from 5/second to 800/second. Useful if you see a model advertised as "30 tokens/second" and want to get a feel for what that actually looks l

Gemini 3.5 Flash: more expensive, but Google plan to use it for everything

9/10

Simon Willison local LLM · May 19 · simonwillison.net

Today at Google I/O, Google released Gemini 3.5 Flash . This one skipped the -preview modifier and went straight to general availability, and Google appear to be using it for a whole lot of their key products: 3.5 Flash is available today to billions of people globally: For every

datasette-llm-accountant 0.1a4

9/10

Simon Willison local LLM · May 19 · simonwillison.net

Release: datasette-llm-accountant 0.1a4 Fixed bug tracking chains of responses. Refs datasette-llm#7 Tags: llm , datasette

By topic

other (1)

v2.1.5

9/10

LiteRT-LM releases local LLM · May 18 · github.com

LiteRT-LM v2.1.5 has been released, introducing Python 3.14 support and making LiteRT C++ APIs header-only. This update also removes the libLiteRt.so dependency from GPU Accelerator and Dispatch API shared libraries, simplifying their usage. Additionally, it adds Raspberry Pi 5 GPU acceleration support and refines the LiteRT Options class for easier manipulation.

This update significantly boosts LiteRT-LM's on-device capabilities, especially with Raspberry Pi 5 GPU acceleration, making it a stronger contender for local LLM deployments and mobile AI projects.

litert-lm on-device ai raspberry pi 5 python 3.14 local llm

Full ranked list

How fast is 10 tokens per second really?

9/10

Simon Willison local LLM · May 20 · simonwillison.net

How fast is 10 tokens per second really? Neat little HTML app by Mike Veerman ( source code here ) which simulates LLM token output speeds from 5/second to 800/second. Useful if you see a model advertised as "30 tokens/second" and want to get a feel for what that actually looks l

Gemini 3.5 Flash: more expensive, but Google plan to use it for everything

9/10

Simon Willison local LLM · May 19 · simonwillison.net

Today at Google I/O, Google released Gemini 3.5 Flash . This one skipped the -preview modifier and went straight to general availability, and Google appear to be using it for a whole lot of their key products: 3.5 Flash is available today to billions of people globally: For every

datasette-llm-accountant 0.1a4

9/10

Simon Willison local LLM · May 19 · simonwillison.net

Release: datasette-llm-accountant 0.1a4 Fixed bug tracking chains of responses. Refs datasette-llm#7 Tags: llm , datasette

llm-gemini 0.32a0

9/10

Simon Willison local LLM · May 19 · simonwillison.net

Release: llm-gemini 0.32a0 Compatible with llm>=0.32a0 alpha - adds the ability to stream reasoning tokens. Tags: gemini , llm

datasette-llm 0.1a8

9/10

Simon Willison local LLM · May 19 · simonwillison.net

Release: datasette-llm 0.1a8 Fix for bug where llm_prompt_context() hook did not fully collect chains of responses. #7

OlmoEarth v1.1: A more efficient family of Earth observation models

9/10

Hugging Face blog local LLM · May 19 · huggingface.co

Fine-Tuning NVIDIA Cosmos Predict 2.5 with LoRA/DoRA for Robot Video Generation

9/10

Hugging Face blog local LLM · May 18 · huggingface.co

The Open Agent Leaderboard

9/10

Hugging Face blog local LLM · May 18 · huggingface.co

v2.1.5

9/10

LiteRT-LM releases local LLM · May 18 · github.com

LiteRT-LM v2.1.5 has been released, introducing Python 3.14 support and making LiteRT C++ APIs header-only. This update also removes the libLiteRt.so dependency from GPU Accelerator and Dispatch API shared libraries, simplifying their usage. Additionally, it adds Raspberry Pi 5 GPU acceleration support and refines the LiteRT Options class for easier manipulation.

This update significantly boosts LiteRT-LM's on-device capabilities, especially with Raspberry Pi 5 GPU acceleration, making it a stronger contender for local LLM deployments and mobile AI projects.

litert-lm on-device ai raspberry pi 5 python 3.14 local llm

An OpenAI model has disproved a central conjecture in discrete geometry

9/10

OpenAI news frontier · May 20 · openai.com

An OpenAI model solved the 80-year-old unit distance problem, disproving a major conjecture in discrete geometry and marking a milestone in AI-driven mathematics.

[AINews] OpenAI GPT-next disproves 80 year old Erdős planar unit distance problem for under $1000

9/10

Latent Space commentary · May 21 · www.latent.space

We will leave coverage of the SpaceXAI IPO filing for the actual day of IPO. Today we celebrate OpenAI’s result, speculated to be GPT 5.6 running for <32 hours or <$1000 , on the planar unit distance problem . Similar to the 2025 IMO Gold result, this is a general purpose LLM, no

[AINews] Google I/O 2026: Gemini 3.5 Flash, Omni (NanoBanana for Video), Spark (background agents), and Antigravity 2.0

9/10

Latent Space commentary · May 20 · www.latent.space

The full keynote livestream was 2 hours, but as usual, The Verge has the best supercut down to 30 mins, which is very worthwhile to get a narrative sense: The mainline Gemini 3.5 Flash is GA today (very nice compared to some staged rollouts) and is sold as a decent step up even c

Huge Codex Upgrade Just Dropped

9/10

Matt Wolfe YouTube commentary · May 18 · www.youtube.com

🚨 OpenAI just dropped the feature we've all been waiting for: Codex on mobile. If you watched my recent video on how I built my custom wiki, you know how much I’ve been relying on Codex. So I was super excited for this feature and I know a bunch of other people have been asking f

llm-gemini 0.32

8/10

Simon Willison local LLM · May 19 · simonwillison.net

Release: llm-gemini 0.32 New model gemini-3.5-flash for Gemini 3.5 Flash . See also my notes on Gemini 3.5 Flash , and the pelican I drew using this upgrade to the plugin. Tags: gemini , llm

The last six months in LLMs in five minutes

8/10

Simon Willison local LLM · May 19 · simonwillison.net

I put together these annotated slides from my five minute lightning talk at PyCon US 2026, using the latest iteration of my annotated presentation tool . # I presented this lightning talk at PyCon US 2026, attempting to summarize the last six months of developments in LLMs in fiv

PaddleOCR 3.5: Running OCR and Document Parsing Tasks with a Transformers Backend

8/10

Hugging Face blog local LLM · May 18 · huggingface.co

A new era for AI Search

8/10

Google AI blog frontier · May 19 · blog.google

We shared the next step in our journey to bring together the best of a search engine with the best of AI.

The Most Important AI News from Google I/O

8/10

AI Daily Brief YouTube commentary · May 20 · www.youtube.com

Google I/O unveiled Omni, Gemini 3.5 Flash, Antigravity 2.0, and Gemini Spark, framing a push toward multimodal generation and agentic tools. Omni showcased powerful video-to-video editing and fine-grained steerability. Gemini 3.5 Flash emphasized speed at the expense of token ef

Introducing the Ettin Reranker Family

7/10

Hugging Face blog local LLM · May 19 · huggingface.co

I/O 2026: Welcome to the agentic Gemini era

7/10

Google AI blog frontier · May 19 · blog.google

The latest from Google I/O: See how we’re helping you get more done with Gemini.

Gemini 3.5: frontier intelligence with action

7/10

Google AI blog frontier · May 19 · blog.google

At Google I/O we released Gemini 3.5, our latest series of models combining frontier intelligence with action.

The next phase of OpenAI’s Education for Countries

7/10

OpenAI news frontier · May 20 · openai.com

OpenAI advances Education for Countries, expanding AI adoption in schools with new partnerships, teacher training, and tools to improve global learning outcomes.

[AINews] How to land a job at a frontier lab (on Pretraining)

7/10

Latent Space commentary · May 19 · www.latent.space

It is the day before Google I/O, when the next major Gemini releases are expected to be previewed, and it will probably be a quiet week from competitors, though Anthropic and OpenAI both had minor wins today, and Cursor shipped their first SpaceXAI model with some nice detail on

The Autonomous Drone Tech Stack & Economics of Drones — Yaroslav Azhnyuk, The Fourth Law & Guest Host Noah Smith, Noahpinion

7/10

Latent Space commentary · May 18 · www.latent.space

The future of war has been evolving before our eyes in Ukraine, yet the west still plans to fight the last war. In this special episode, guest host ( @noahpinion ) and sit down with Yaroslav Azhnyuk ( @YaroslavAzhnyuk ) , a serial tech founder who went from building PetCube to fo

Viral Post Embarrassingly Exposes AI Haters 💀

7/10

Matt Wolfe YouTube commentary · May 20 · www.youtube.com

Internet art critics just took the biggest L of the year 💀 A user on X posted a REAL Monet painting, slapped an "AI Generated" tag on it, and asked people to critique it. People wrote whole essays about how the painting was "soulless," "lacked human touch," and "obviously a compu

Krea AI Launches Crazy New Image Model

7/10

Matt Wolfe YouTube commentary · May 19 · www.youtube.com

How to get the PERFECT AI art style every single time: 1. Go to Krea.ai and toggle to the "Krea 2 Large" model for photorealism or "Krea 2 Medium" model for illustrations 2. Click "Mood Boards" on the bottom dashboard 3. Drop in a batch of images that share the aesthetic you want

9 Codex Tips from the Codex Team

7/10

AI Daily Brief YouTube commentary · May 20 · www.youtube.com

Composer 2.5 narrows the gap with frontier coding models on key benchmarks while Cursor touts dramatic token‑efficiency at a fraction of the cost. Enterprise strategy is shifting toward harness‑first platforms and agent orchestration, capturing long‑running context, persistent me

What Google Needs to Do at I/O This Week

7/10

AI Daily Brief YouTube commentary · May 18 · www.youtube.com

Codex integration into ChatGPT Mobile ushers in persistent agent workflows and shifts knowledge work toward human review and approval instead of direct execution. Google I/O expectations include a Gemini Spark personal agent and cost-optimized Gemini Flash models pursuing both co

Google I/O, Gemini Spark, Antigravity

6/10

Simon Willison local LLM · May 20 · simonwillison.net

It's hard to find much to write about Google I/O this year because I have a policy of not writing about anything that I can't try out myself, and a lot of the big announcements are "coming soon". I actually prefer to write about things that are in general availability, because I'

How Ramp engineers accelerate code review with Codex

6/10

OpenAI news frontier · May 20 · openai.com

How Ramp engineers use Codex with GPT-5.5 to review code and ship improvements, allowing them to get substantive feedback in minutes instead of hours.

Introducing OpenAI for Singapore

6/10

OpenAI news frontier · May 19 · openai.com

OpenAI for Singapore launches a multi-year AI partnership to expand deployment, build local talent, and support businesses and public services with AI.

Advancing content provenance for a safer, more transparent AI ecosystem

6/10

OpenAI news frontier · May 19 · openai.com

OpenAI advances AI content provenance with Content Credentials, SynthID, and a verification tool to help people identify and trust AI-generated media.

Railway: The Agent-Native Cloud — Jake Cooper

6/10

Latent Space commentary · May 20 · www.latent.space

Take the 2026 AI Engineering Survey and get >$2k in credits and AIE WF tickets ! This was recorded before Railway suffered a major GCP outage on May 19, despite being a multi-AZ, multi-zone mesh ring, with HA fiber interconnects between their Metal <> GCP <> AWS, because workload

I/O 2026

5/10

Google AI blog frontier · May 19 · blog.google

At Google I/O 2026, we shared how we’re making AI more helpful for everyone. See everything we announced.

How AI Mode is changing the way people search in the U.S.

5/10

Google AI blog frontier · May 19 · blog.google

One year after launch, see how AI Mode’s users are shifting from keywords to natural language queries.

New ways to create and get things done in Google Workspace

5/10

Google AI blog frontier · May 19 · blog.google

Announcing new voice capabilities in Gmail, Docs and Keep, a new design tool called Google Pics and updates to AI Inbox.

OpenAI and Dell partner to bring Codex to hybrid and on-premise enterprise environments

5/10

OpenAI news frontier · May 18 · openai.com

OpenAI and Dell partner to bring Codex to hybrid and on-premise environments, helping enterprises deploy AI coding agents securely across data and workflows.

Pipeline status

Pipeline stats

Dead sources today

All sources green today.