GPT\u20116 Astra is finally live, but if you\u2019re refreshing ChatGPT hoping to see a shiny \u201cAstra\u201d button, you\u2019re not alone in being confused. OpenAI launched the model on September 3, then rolled it out in stages after what Sam Altman himself called a \u201cmessy\u201d start.<\/p>\n\n\n\n
On the consumer side, Pro ($100 and $200 tiers), Business and Enterprise<\/strong> users are seeing GPT\u20116 Astra appear in the model picker on web, iOS and Android, with Pro also getting access to an Astra Pro<\/a><\/strong> variant for heavier workloads. Plus subscribers<\/strong> are getting Astra mainly inside ChatGPT Work and Codex<\/strong>, not the standard chat, and with limited usage before credits kick in. Free and Go users are still excluded from Astra entirely for now.<\/p>\n\n\n\n For developers, the model is available via the OpenAI API<\/strong> as On the API, OpenAI\u2019s standard pricing for GPT\u20116 Astra is $10 per million input tokens<\/strong> and $50 per million output tokens<\/strong>, with cached inputs at $1 per million<\/strong> and cache writes at $12.50 per million<\/strong>. There\u2019s also a fast mode<\/strong> that runs up to 2\u00d7 speed at 2\u00d7 the standard rate<\/strong>.<\/p>\n\n\n\n Inside ChatGPT subscriptions, Astra usage is bundled into your existing Work\/Codex allowances for Pro, Business and Enterprise, with extra usage charged as credits; Plus gets a smaller included slice before overage charges apply.<\/p>\n\n\n\n OpenAI is positioning Astra as its computer\u2011use and coding flagship<\/strong>, not just a better chat model. On Terminal\u2011Bench 4.0<\/strong>, Astra scores around 58.18%<\/strong>, ahead of GPT\u20115.6 Sol (~37\u201338%) and Claude Fable 5.1<\/a> (~55\u201356%). On OSWorld 2.0<\/strong>, Astra hits 72.6%<\/strong> versus Sol\u2019s 65.7%<\/strong>, and on ScreenSpot\u2011Pro<\/strong> it reaches 92.7%<\/strong> compared to Sol\u2019s 76.9%<\/strong>.<\/p>\n\n\n\n Astra shines on long, multi\u2011step terminal work, browser and desktop automation, and complex coding agents, while older models still handle simple Q&A just fine. If your workflow leans heavily on agents, code and computer control, Astra is the one OpenAI<\/a> wants you running.<\/p>\n","protected":false},"excerpt":{"rendered":"gpt-6-astra<\/code>, plus through Microsoft Azure<\/strong> and AWS Bedrock<\/strong> integrations. Enterprise workspaces can enable Astra, but it\u2019s often off by default until admins turn it on and set policies.<\/p>\n\n\n\nGPT-6 Astra Pricing: what you\u2019ll actually pay<\/h2>\n\n\n\n
GPT-6 Astra Benchmarks<\/h2>\n\n\n\n