ai-models - The LGTM (Page 3)

The LGTM

Home
Agentic Coding
Claude Code
Codex

Sign in Subscribe

ai-models

A collection of 47 posts

GitHub Copilot's June 1 Billing Reset: What Usage-Based Pricing Actually Costs You

On June 1st, GitHub flips a switch that every Copilot subscriber needs to understand before they open their June bill. The change is simple on paper: GitHub Copilot moves from flat-rate, request-based billing to per-token usage-based pricing. Every Copilot Pro ($10/month), Copilot Pro+ ($19/month), Business, and Enterprise subscriber

01 May 2026 5 min read

Musk Just Got Caught in the Most Damaging Cross-Examination of the AI Decade

Musk Just Got Caught in the Most Damaging Cross-Examination of the AI Decade

On Thursday in a California federal courtroom, Elon Musk was asked a simple question under oath: has xAI used OpenAI's models to train Grok? He paused, shifted, and said "Partly." That single word is the most consequential admission to come out of the AI industry'

30 Apr 2026 4 min read

OpenAI's Restricted Cyber Model Follows the Same Playbook It Mocked in Others

OpenAI's Restricted Cyber Model Follows the Same Playbook It Mocked in Others

Two weeks ago, Sam Altman called Anthropic's decision to gatekeep its Claude Mythos cyber model "fear-based marketing." Today, OpenAI confirmed it is doing exactly the same thing with GPT-5.5-Cyber — restricting access to vetted "critical cyber defenders" under a government-coordination framework, with an application

30 Apr 2026 4 min read

Mistral Medium 3.5's Real Launch Feature Is the 70GB VRAM Trick, Not the Benchmark Score

Mistral dropped Mistral Medium 3.5 today, and the benchmark chart is doing the usual rounds. 77.6% on SWE-Bench Verified, 91.4% on τ³-Telecom, competitive with Qwen3.5 397B A17B per Mistral's own numbers. Fine. That is the press release version, and you have already seen

29 Apr 2026 5 min read

Nemotron 3 Nano Omni Is NVIDIA's Most Aggressive Move Into the Model Business Yet

Nemotron 3 Nano Omni Is NVIDIA's Most Aggressive Move Into the Model Business Yet

NVIDIA has been careful for years to frame its open models as complements to proprietary frontier models rather than direct competitors. Nemotron 3 Nano Omni, unveiled April 28, continues that pattern while making the complement story a lot harder to dismiss. The model is a 30B-A3B hybrid MoE that activates

28 Apr 2026 5 min read

DeepSeek V4 Is the First Open Model Release This Year That Actually Treats Context Length Like an Engineering Constraint

DeepSeek V4 Is the First Open Model Release This Year That Actually Treats Context Length Like an Engineering Constraint

Open models keep advertising giant context windows the way cloud vendors advertise unlimited storage: technically true, operationally slippery. The hard part is not printing 1M tokens on a model card. The hard part is building a model and serving stack that do not melt once an agent spends an afternoon

24 Apr 2026 5 min read

OpenAI’s GPT-5.5 Bio Bug Bounty Is a Quiet Admission That Model Safety Needs Adversaries, Not Just Policies

OpenAI’s GPT-5.5 bio bug bounty is a small announcement with a large implication. The company is offering $25,000 to the first vetted researcher who can find a universal jailbreak that defeats its five-question biology safety challenge in GPT-5.5 from a clean chat in Codex Desktop. On

23 Apr 2026 4 min read

GPT-5.5 Looks Like the First Frontier Launch Optimized for Real Work, Not Just Benchmark Theater

OpenAI did not launch GPT-5.5 as a pure benchmark flex. It launched it as an argument about labor economics. The interesting claim in the company’s release is not that one more flagship model beat last quarter’s flagship on a grid of evals. That is table stakes now.

23 Apr 2026 4 min read

The GPT-5.5 System Card Says Frontier Models Are Crossing Into Capability Management, Not Just Capability Marketing

The most useful OpenAI document published with GPT-5.5 is not the launch post. It is the system card, because that is where the company quietly admits what frontier AI deployment has become. We are no longer in the era where the main question is whether a model can clear

23 Apr 2026 4 min read

GPT-5.2’s Science Pitch Is Really a Reliability Pitch

OpenAI keeps trying to talk about science without sounding like it is promising a robot Nobel Prize. For once, that restraint is the interesting part. The new GPT-5.2 science and math note reads less like a moonshot manifesto and more like a product manager finally admitting what researchers actually

22 Apr 2026 4 min read

Gemma 4 on Cheap Edge Hardware Is a Better Model Story Than Another Leaderboard

Gemma 4 on Cheap Edge Hardware Is a Better Model Story Than Another Leaderboard

The most useful AI model story of the night is not a new frontier release, and that is exactly why it matters. A same-day Hugging Face post from NVIDIA engineer Asier Arranz shows a multimodal Gemma 4 setup running locally on an 8 GB Jetson Orin Nano Super, complete with

22 Apr 2026 5 min read

ImageGen 2.0 Shows OpenAI Is Turning Image Models Into Routed Compute Tiers, Not Just a Magic Button

ImageGen 2.0 Shows OpenAI Is Turning Image Models Into Routed Compute Tiers, Not Just a Magic Button

OpenAI’s ImageGen 2.0 update looks small if you read it like a release note and much more important if you read it like product architecture. The company added a baseline ImageGen 2.0 tier to ChatGPT for all plans, then introduced ImageGen 2.0 Thinking for paid users,

21 Apr 2026 5 min read

← Newer Posts Page 3 of 4 Older Posts →

The LGTM © 2026

Sign up

Powered by Ghost