{
  "id": 6146411,
  "title": "OpenAI’s new model costs 2.5x more per token — and developers are saving money anyway",
  "url": "https://urgent.news/2026/09/07/openais-new-model-costs-2-5x-more-per-token-and-developers-are-saving",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-07T16:12:18.000Z",
  "source": {
    "name": "The New Stack",
    "slug": "the-new-stack",
    "url": "https://thenewstack.io/astra-reasoning-effort-cost/"
  },
  "original_language": "en",
  "account": "OpenAI's newest model, GPT-6 Astra, has higher token costs compared to its predecessor, GPT-5.6 Sol. However, developers may still opt to upgrade and adjust the reasoning setting, as Astra can outperform Sol at lower reasoning levels. Astra's low reasoning level performs better than Sol's high level, and Astra's medium level often provides a cost-effective alternative. In a comparison test, Astra-medium completed tasks more quickly and used fewer input tokens than Sol-high, resulting in a lower estimated cost. While more reasoning may lead to better results in some cases, the additional cost may not always be justified. Astra also offers a configuration_update mechanism, allowing developers to change reasoning levels dynamically during a single request. This flexibility enables a more cost-efficient workflow, as routine tasks can use lower reasoning levels, and complex problems can trigger higher reasoning when needed. Ultimately, the cost-effectiveness of Astra depends on the specific workload and the balance between reasoning level and token usage.",
  "summary": "GPT-6 Astra has an obvious problem for developers considering an upgrade from GPT-5.6 Sol: its tokens cost 2.5 times as The post OpenAI’s new model costs 2.5x more per token — and developers are saving money anyway appeared first on The New Stack .",
  "key_points": [
    "GPT-6 Astra has higher token costs than GPT-5.6 Sol.",
    "Astra's low reasoning level outperforms Sol's high level at lower reasoning.",
    "Astra's configurationupdate allows dynamic reasoning level changes."
  ],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "PC Gamer",
        "title": "Jensen Huang says 100,000 Nvidia GPUs were used to train OpenAI's latest model, GPT-6 Astra, and there's already plans to bring quadruple that amount of hardware online",
        "url": "https://urgent.news/2026/09/07/jensen-huang-says-100-000-nvidia-gpus-were-used-to-train-openais",
        "published": "2026-09-07T15:20:28.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}