Skip to content

Greg Brockman Says GPT-6.2 Is Incredible at Browser Use

Greg Brockman · XRelayed· @gdb·· 1 day agoPicks

21.6K views · 513 likes · 21 reposts (at collection)

Original

What happened

Greg Brockman says GPT-6.2 is incredible at browser use and quotes another account calling GPT-6.2 Sol the top model for the task. The quoted post reports an 82% score, up from 81% for GPT-6.1 Sol, at 35% lower cost, twice the speed, and 36% fewer steps.

Why it matters

In the cited browser use comparison, GPT-6.2 Sol raises the score from 81% to 82% while costing 35% less and taking 12 minutes instead of 22.

Article

GPT-6.2 is incredible at browser use:

Quoted postBrowser Use@browser_use

GPT-6.2 Sol is the top model for browser use 👀

82%, up from 81% for GPT-6.1 Sol
35% cheaper
2x faster: 12 min instead of 22 min
36% fewer steps

View the quoted post on X

The story so far

OpenAI released GPT-6.2 Astra and Sol, which are now available in ChatGPT Work, Codex, and the API. Astra features a 20% reduction in API input and output token pricing compared to its predecessor, while Sol maintains the pricing structure of GPT-6.1 Sol. The company also expanded access to Ultrafast mode for Pro 500 subscribers using either model and for Pro 200 subscribers using Sol, with a temporary 50% price cut for API and Enterprise users. OpenAI reported that the new models set performance records on benchmarks like Agents' Last Exam and BenchCAD 2.1, and Perplexity observed gains on its WANDR benchmark after integrating both models. However, feedback has been mixed. Independent evaluator Simon Willison noted the models struggled with animation tasks, and Epoch AI reported that GPT-6.2 Astra trailed Fable 5.2 on internal automation quality standards. Additionally, Jeremy Howard of Solveit reported reverting to versions of Astra and Sol prior to 6.2, stating the update was not working for him or his users. OpenAI simultaneously implemented tighter cybersecurity safeguards, which the company noted may increase refusal rates for some security research tasks.

How the story developed · 4

View full event
  1. 11/19 00:33 (UTC+8)

    Greg Brockman Says GPT-6.2 Is Incredible at Browser Use This report

  2. 11/18 06:15 (UTC+8)
    Tibor Blaho Tests GPT-6.2 Astra and Sol on Daily WorkTibor Blaho · X
  3. 11/18 06:15 (UTC+8)
    Perplexity Shares GPT-6.2 Astra and Sol WANDR EvalsAravind Srinivas · X
  4. 11/18 02:36 (UTC+8)
    OpenAI Introduces GPT-6.2 Astra and SolOpenAI News

Read next

See the full storyOpenAI Launches GPT-6.2 Astra and Sol, With Mixed Performance and Recent Version Reversions
This story is in 11/19 AI DailyRead the whole issue