Model intelligence.
Follow the changes. Understand what comes next.
Google Launches Gemini 4 Neon in Antigravity With 1M Context, Four Thinking Levels, and Benchmarks vs. Claude Opus 5.5 and GPT-6.2 Sol
Announced on November 19, 2026 at 05:51 Beijing time, Google's Gemini 4 Neon features a 1M-token context window, four thinking levels, and planner-executor pairing with Gemini 4 Hydrogen in Google Antigravity, outperforming Claude Opus 5.5 and GPT-6.2 Sol across most of 18 published benchmarks.
Latest stories349
Newest publishedReflection AI releases Beam 1.0, a 501B open coding model under Apache 2.0
Reflection AI has published the weights for Beam 1.0, a 501B mixture-of-experts model with 23B active parameters and a 1M-token context, under Apache 2.0. It also open-sourced Mirror CLI, the coding harness it trained Beam on.
OpenCode makes GLM 5.3 Flash free for a limited time
OpenCode now offers GLM 5.3 Flash as a free Console model under the ID glm-5.3-flash-free. The docs call it a limited-time free tier with zero retention and no training on your data.
Nemotron 3.5 Super VL and Bolt: Specs, License, Runs on 24GB
NVIDIA released Nemotron 3.5 Super VL (120B MoE, 12B active, 1M context, vision) and Bolt 74B (from Super VL, 9B active, native 2-bit, ~24GB, still images/video) on October 15 under OpenMDW-1.1. On October 23 it added Bolt E4B—post-trained from Gemma 4 E4B, 8B total / 4B effective, text in and text out, 64K context. The “4GB” in the name is NVIDIA’s target memory envelope for its workloads; that is not the Q4_K_M weight file size (~6.18GB), and E4B is not a shrink of the 74B Bolt.
MiniCPM-V 4.7 puts a 35B open multimodal model on phones, using about 3GB of memory
ModelBest and OpenBMB released MiniCPM-V 4.7 in 35B-A3B and 1B sizes under Apache-2.0. The company says the 35B MoE model activates about 3B parameters per token and runs on iPhone and iPad in about 3GB of memory.
Grok Voice 3.0 tops the Artificial Analysis speech-to-speech index, at the same $0.08 a minute
xAI's new speech-to-speech model scores 86.9 on Artificial Analysis's Speech-to-Speech Index, first place by xAI's account. The Voice Agent API price stays at $0.08 per minute, and agents on grok-voice-latest have already been switched over.
Gemini 4 Argon is out globally — $2/$10 now, $4/$20 after November 30
Google opened Gemini 4 Argon to paid API users and to Ultra and Pro subscribers in the Gemini app, Search AI Mode, and Antigravity. Intro pricing is $2 input / $10 output per million tokens until November 30, then $4 / $20. Context is a unified 1M tokens.
GPT-6.1 Sol drops the long-context fee on subscriptions — and the default context jumps to 500k
In Codex and ChatGPT Work, GPT-6.1 Sol no longer adds a long-context fee when usage comes from your subscription. The default context also rises from 272k to 500k tokens. API pricing above 272K input is unchanged.
Devin can bill GPT usage to your personal ChatGPT plan — the rest still draws from Devin
Cognition lets you link ChatGPT Go, Plus or Pro to Devin Pro, Max or Teams. GPT calls (except fast and priority) then come out of your ChatGPT quota; other models still use Devin. When ChatGPT runs out, Devin falls back to its own quota, then on-demand credits.
ChatGPT Dot can now start work in Codex and follow up on your threads
Your ChatGPT Dot can open new Codex work and pick up existing threads, using your ChatGPT chats, Codex threads and automations to decide whether to continue or start fresh. The ChatGPT account posted it on X; Tibo shared the same upgrade later the same day.
Codex Composer now predicts your next message for Pro users — and generating it doesn't burn usage
Codex on the desktop can now suggest your next message for personal Pro users, based on the conversation and how you talk to it. OpenAI says it's in beta, and generating a suggestion doesn't count toward usage. Windows users also get a new MXC sandbox that needs a compatible Windows 11 build.
ByteDance's TRAE merges TraeCode and TraeWork into one app with Agent and IDE modes
TRAE has folded its coding app and its work assistant into a single TRAE with an Agent mode for running several agents in parallel and an IDE mode for hands-on editing. TraeWork shuts down on October 23, so existing users need to upgrade before then.
Cursor Remote Control lets you check on and reply to your local agents from an iPhone
Cursor's iOS app can now connect to the agents running on your own computer, so you can see what they're doing and send them messages from your phone. It's on by default except for Enterprise, it only works while your computer stays on, and Cursor hasn't said anything about Android.