HaiAI123

Curated Global AI Tools Directory

RisingIndustry update
↑ 7%

GPT-6.2 Astra and Sol now generally available, with deployment comparisons

11/21/2026 — 11/22, 07:17·5 sources (all)·7 reports (all)·In the 2026-11-22 briefing

Key takeaways

  • OpenAI's GPT-6.2 Astra and GPT-6.2 Sol became generally available on November 17, 2026, with Astra upgrading GPT-6 Astra and Sol building on…

Background & analysis

On November 17, 2026, OpenAI made both GPT-6.2 Astra and GPT-6.2 Sol generally available. GPT-6.2 Astra is the upgrade to GPT-6 Astra, while GPT-6.2 Sol builds on GPT-6.1 Sol. That same day, OpenAI cut the per-token price of GPT-6.2 Astra by 20%. On November 20, the company published a piece describing its layered approach to adversarial distillation, combining model training, safeguards, detection, and enforcement to prevent capability extraction, and framed the issue as an ongoing back-and-forth requiring both training-side and detection-side measures rather than a single technical fix.

On November 21, several outlets added detail to that announcement. According to IT Home, OpenAI said it had strengthened protection for the internal reasoning Chain-of-Thought in GPT-6.2 Astra and GPT-6.2 Sol. Its internal adaptive black-box red team testing aggregated data from 6 attack models, cutting the extraction success rate for previously requested reasoning content from 54.0% to 0.2%, and for currently requested content from 22.9% to 0.9%. OpenAI defined adversarial distillation as training other models on outputs from frontier models without authorization, giving them similar capabilities without the original model's safeguards. A DEV Community report tied the 54.0% to 0.2% figure to a copying attack that tries to pull out hidden reasoning, and noted OpenAI had earlier linked such attempts to DeepSeek and people associated with Kimi maker Moonshot AI. QbitAI reported that GPT-6.2 Astra doubles GPT-6 Astra's success rate on 16 embodied tasks and claimed a first-ever pass in deformable object manipulation, without publishing the task list, test conditions, or evaluation method.

Two more pieces on the same day focused on deployment. A DEV Community analysis cited evaluator Cognition, which found GPT-6.2 Sol within half a point of GPT-6.2 Astra on a major coding benchmark at roughly one-fifth of the cost per task, arguing the gap between the high-priced and low-priced models is shrinking. A comparison aimed at C# developers offered a sample program using C# and Microsoft.Extensions.AI to send the same task to both models, while cautioning that the available material contained only Foundry testing-style claims and no benchmark numbers. The Decoder reported that OpenAI released new prompt templates for GPT-6.2 Astra and Sol to fix overly brief answers, unnecessary follow-up questions, and excessive testing when coding, and warned that editing conversation history in the API can discard reasoning content and lower response quality. Public reporting currently stands at the disclosure stage of model availability, pricing changes, safety test figures, and deployment comparisons, with no further developments reported.

AI-generated from 7 reports · updated 3 hours ago

Latest turnGPT-6.2 Astra and GPT-6.2 Sol both became generally available on November 17, 2026, with Astra upgrading GPT-6 Astra and Sol built on GPT-6.1 Sol, according to the post. It compares the stated changes and per-task pricing and includes a .NET 10 sample using Microsoft.Extensions.AI that sends the same work to both models, while noting the available material offers positioning rather than benchmark numbers.

Related AI tools

updream
24-hour heatHeat index 290 · peak 391 · 12h ago
24 hours agonow

Reports on this story headlines open the original

Today
  1. GPT-6.2 Astra and GPT-6.2 Sol both became generally available on November 17, 2026, with Astra upgrading GPT-6 Astra and Sol built on GPT-6.1 Sol, according to the post. It compares the stated changes and per-task pricing and includes a .NET 10 sample using Microsoft.Extensions.AI that sends the same work to both models, while noting the available material offers positioning rather than benchmark numbers.

    DEV Community · AIAI score 53
Yesterday
  1. OpenAI has released new prompt templates for GPT-6.2 Astra and Sol to address overly brief answers, unnecessary follow-up questions, and excessive testing in coding tasks. The Decoder notes an API change: editing a conversation's history can discard reasoning content and lower response quality.

    The DecoderAI score 65
  2. OpenAI cut the per-token price of GPT-6.2 Astra by 20% on November 17, but a DEV Community analysis argues the bigger shift sits one tier down: GPT-6.2 Sol landed within half a point of Astra on a major coding benchmark, according to evaluator Cognition, at roughly a fifth of the cost per task. The price cut, the author argues, is not the real story.

    DEV Community · AIAI score 55
  3. According to QuantumBit, GPT-6.2 Astra doubles the success rate of GPT-6 Astra across 16 embodied tasks and claims the first successful deformable-object manipulation of its kind. The report does not detail the task list, test conditions, or evaluation setup.

    量子位AI score 60
  4. OpenAI published a note on its defenses against adversarial distillation, saying it hardened chain-of-thought protection in GPT-6.2 Astra and GPT-6.2 Sol, IT之家 reports. In internal adaptive black-box red-team tests across six attack models, extraction of protected reasoning from prior requests fell from 54.0% to 0.2%, and from 22.9% to 0.9% for current requests.

    IT之家AI score 56
  5. OpenAI says a copying attack that tries to extract a model's hidden reasoning now succeeds only 0.2% of the time on GPT-6.2 Astra, down from 54%. The company had earlier linked such attempts to China's DeepSeek and people tied to Kimi maker Moonshot AI.

    DEV Community · AIAI score 63
Nov 20
  1. OpenAI published an overview of its approach to adversarial distillation, saying it combines model training, safeguards, detection and enforcement to protect against capability extraction. The company frames the problem as an ongoing, layered effort rather than something a single technical fix can address.

    OpenAI NewsFirst-partyAI score 53

Other stories people are talking about

How is heat calculated?About the method

Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.

This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.

Surge
Discussion rising fast
New
First report within 6 hours
Rising
Still gathering discussion

Back to AI industry updates →