<!-- Search description: OpenAI cut GPT-6.2 Astra API prices 20% to $8 per million input tokens at launch. What the GPT-6.2 Astra and Sol models mean for AI costs. -->
<h2>How much does GPT-6.2 Astra cost in the API?</h2>
<p style="text-align: justify;">OpenAI released its new flagship model, GPT-6.2 Astra, and a cheaper everyday model, GPT-6.2 Sol, on Tuesday, November 17, 2026, and priced the launch as aggressively as it sold the performance: Astra's standard API rate fell 20 percent to $8 per million input tokens and $40 per million output tokens, while cached input dropped 60 percent to $0.40. Sol keeps the $2 input and $10 output rates GPT-6.1 Sol already carried. Both models are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex and through the API, and Microsoft's GitHub made them available in Copilot for Pro+, Max, Business and Enterprise users the same day.</p>
<h2>What is the difference between GPT-6.2 Astra and GPT-6.2 Sol?</h2>
<p style="text-align: justify;">Astra is OpenAI's frontier model for long, complex work: financial analysis, legal drafting and hard programming. Sol is the efficient workhorse for everyday and multistep agentic tasks. The pricing gap is the point: OpenAI says Sol gets within half a point of Astra on one coding benchmark at roughly one-fifth of the cost, and outscores the previous flagship GPT-6 Astra on its everyday-task LifeVal test at about one-seventh of the cost per task. Subscribers on the Pro 200 plan get the faster Ultrafast mode for the first time, starting with Sol, and Pro 500 subscribers can use it with either model.</p>
<h2>Why did OpenAI cut the flagship's prices?</h2>
<p style="text-align: justify;">The cut is aimed at a market where rivals now compete on cost per finished task, not just raw scores. Anthropic's Claude Fable 5.2 charges the same $8 and $40 headline rates Astra moved to, and Google's Gemini 4 Neon launched as a deliberately cheaper mid-tier option at $3 per million input tokens. On OpenAI's own numbers, the new Astra scored 6.2 percentage points higher than GPT-6 Astra on the GDPval v2 benchmark of real professional tasks while using 22 percent fewer output tokens, and beat its predecessor by 5.2 points on Agents' Last Exam at under a third of the cost per task. Lower rates only pay off if usage grows, which is exactly the bet the company is making.</p>
<h2>Do independent testers agree Astra is the best model for scientists?</h2>
<p style="text-align: justify;">OpenAI calls GPT-6.2 Astra the best model in the world for scientists. Early testers are cooler on the claim. The independent benchmarker Artificial Analysis scored Astra 62 on its Intelligence Index, behind Claude Fable 5.2 at 65, and the research group Epoch AI found Astra trailed Fable 5.2 on realistic work tasks. Developer Theo Browne wrote that he would not choose it over Anthropic's Opus 5.5, and Wharton professor Ethan Mollick called the model "very unstable and very weird" even while praising a 3D space-elevator demo it built. Dan Shipper, chief executive of the newsletter Every, noted that Astra often stops halfway through coding tasks. OpenAI's answer to the criticism: many of its own employees have switched to the cheaper Sol as their default for everyday and technical work.</p>
<h2>What does the price cut mean for AI builders?</h2>
<p style="text-align: justify;">For teams pricing out AI features, the two changes that matter are the 20 percent headline cut and the 60 percent discount on cached input, the latter a direct saving for applications that reuse the same long context, such as document sets and codebases. Both models meet OpenAI's Critical threshold for cybersecurity capability and ship with tighter limits on higher-risk cyber requests; authorised defenders can still use GPT-6 Cyber through the company's Daybreak programme. The practical takeaway for Nigerian startups and any budget-conscious builder is simple: the cost per useful unit of AI work is falling fast enough that last quarter's model budget is worth repricing.</p>
<h3>FAQ</h3>
<p style="text-align: justify;"><strong>How much does GPT-6.2 Astra cost?</strong> $8 per million input tokens and $40 per million output tokens in the API, with cached input at $0.40, a 20 percent cut from GPT-6 Astra's rates.</p>
<p style="text-align: justify;"><strong>Is GPT-6.2 Sol cheaper than Astra?</strong> Yes. Sol keeps GPT-6.1 Sol's pricing at $2 per million input tokens and $10 per million output tokens.</p>
<p style="text-align: justify;"><strong>When did GPT-6.2 Astra and Sol launch?</strong> Tuesday, November 17, 2026, in ChatGPT Work, Codex, the developer API and GitHub Copilot.</p>
<p style="text-align: justify;"><em>Related: <a href="https://www.bhginfodesks.com/2026/11/openai-launches-gpt-6-cyber-its-first.html">OpenAI Launches GPT-6 Cyber, Its First Model Built For Hacking And Cyber Defence</a> and <a href="https://www.bhginfodesks.com/2026/11/microsoft-rolls-out-new-mai-models.html">Microsoft Rolls Out New MAI Models: Cyber 2 Flash, Code 2, Image 2.7 And Dragon Copilot Transcribe</a>.</em></p>
<p style="text-align: justify;"><em>Image: A woman walks between glowing server racks in a data centre, a look at the compute that pays the API bill behind every model launch. Photo: franganillo via Pixabay.</em></p>
0 Comments