Robots Atlas>ROBOTS ATLAS

Tag: Gpt 6 astra

Gemini 4 Carbon: 88.2% at finding flaws, but no access yet
Aktualnoƛci11 listopada 2026

Gemini 4 Carbon: 88.2% at finding flaws, but no access yet

Google DeepMind released Gemini 4 Carbon two weeks after Argon. It set records on three cybersecurity benchmarks and repelled every evaluated prompt injection, but access is limited to selected cyber defence teams for now.

Astra disabled its monitor, then denied it: OpenAI's report
Aktualnoƛci29 paĆșdziernika 2026

Astra disabled its monitor, then denied it: OpenAI's report

OpenAI described two incidents in which coding agents weakened their own misalignment monitoring. Astra removed the monitoring integration and denied it when the user asked directly.

GPT-6 lands on Google Cloud, three closed LLMs at one provider
Aktualnoƛci29 paĆșdziernika 2026

GPT-6 lands on Google Cloud, three closed LLMs at one provider

Google Cloud launched OpenAI models in its model garden: GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna, for now in private preview. Billing runs through Google Cloud Marketplace, and Microsoft remains OpenAI's primary cloud partner.

Gemini 4 Argon goes global: 68.9% on Vals Index, $2 per million tokens
Aktualnoƛci25 paĆșdziernika 2026

Gemini 4 Argon goes global: 68.9% on Vals Index, $2 per million tokens

Argon is globally available through the API, in Antigravity and in the Gemini app. Google shows a Chromium refactor and a 2.85× Gemma 4 speedup, and the Vals Index puts the model at 68.90%.

Gemini 4 Argon: a million output tokens and 77.9% on DeepSWE
Aktualnoƛci15 paĆșdziernika 2026

Gemini 4 Argon: a million output tokens and 77.9% on DeepSWE

Google DeepMind introduced Gemini 4 Argon on 30 September 2026. The generation limit rises from 64,000 to one million tokens, and the model reaches cyber defenders before general developers.

Opus 5.5 dropped the em-dashes. 2,548 tells remain
Aktualnoƛci14 paĆșdziernika 2026

Opus 5.5 dropped the em-dashes. 2,548 tells remain

Graphite analysed 9,974 topics across ten models. Opus 5.5 cut em-dash use by 99%, yet its count of overused phrases fell only 4%.

GPT-6.1 Sol: Astra-level work at one-fifth of the price
Aktualnoƛci4 paĆșdziernika 2026

GPT-6.1 Sol: Astra-level work at one-fifth of the price

OpenAI priced GPT-6.1 Sol at exactly one-fifth of GPT-6 Astra's rates, with benchmark gaps of about two percentage points. The new Ultrafast tier delivers 300 tokens per second at six times the price.

OpenAI launches Dots: agents that work after you close the chat
Aktualnoƛci4 paĆșdziernika 2026

OpenAI launches Dots: agents that work after you close the chat

At DevDay 2026 OpenAI unveiled Dots, GPT-6 Astra agents that keep working after the chat closes, plus ChatGPT Space for humans and agents. Limits and pricing stay undisclosed.

GPT-6 Astra drove a real car. The only model of four to finish
Aktualnoƛci1 paĆșdziernika 2026

GPT-6 Astra drove a real car. The only model of four to finish

DrivingBench, an independent project, gave four frontier language models control of a real Toyota Corolla. Only GPT-6 Astra finished the parking-lot cone course: 134.7 metres in 5 minutes and 22 seconds, on its second attempt.

GPT-6 Astra Drives a Robot Arm — but Only on the Easy Task
Aktualnoƛci29 wrzeƛnia 2026

GPT-6 Astra Drives a Robot Arm — but Only on the Easy Task

GPT-6 Astra finished 19 of 20 block-moving runs against 8 for Claude Fable 5.1, using six times fewer tokens. On the precision task both score 2/20, and RoboDojo halted testing after equipment damage.