OpenAI Launches GPT-6.2 Astra and Sol, Cutting Frontier Model Costs by 20%

  • Dispatched: Nov 18, 2026 · 18:18 ET (23:18 UTC)

One-sentence summary: OpenAI on 2026-11-17 12:45 ET / 17:45 UTC launched GPT-6.2 Astra and GPT-6.2 Sol: the flagship Astra beats its predecessor by 5.2 points on a multistep professional-work benchmark at less than a third of the cost, and beats Anthropic's Claude Fable 5.2 by 5.5 points at just 5% of the cost, with API prices 20% below GPT-6 Astra.

Verification status: ✓ Official first-party disclosure (OpenAI official blog and official thread); launch confirmed the same day by Reuters.

Core Facts & Consensus Delta:

  • Launch: OpenAI official blog, 2026-11-17 12:45 ET / 17:45 UTC; official thread at 12:54 ET / 17:54 UTC. Astra is the frontier model for the hardest professional work (slides, spreadsheets, documents, code, frontends, 3D and game design); Sol is the efficient, versatile model for everyday work.
  • API pricing (per million tokens): Astra $8 input, $0.40 cached input, $40 output — 20% below GPT-6 Astra, with a 60% discount on cached input. Sol keeps GPT-6.1 Sol pricing ($2 input, $0.10 cached, $10 output).
  • Benchmarks (OpenAI's own figures): On Agents' Last Exam, Astra scores 5.2 points above GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Claude Fable 5.2 at just 5% of the cost (Fable 5.2's leaderboard score excludes nine tasks that used a fallback model). On Terminal-Bench-Science 0.1, Astra scores 9 percentage points above its predecessor at less than a third of the cost. Artificial Analysis Intelligence Index: Astra 62 (up from 57), second only to Claude Fable 5.2 at 65; Sol 57.3 (up from 55).
  • Safety ratings: Both models meet the Critical threshold for cybersecurity under OpenAI's Preparedness Framework, launching with tighter safeguards on higher-risk cybersecurity assistance. Astra is the first externally released model to meet the High threshold for AI self-improvement; both models also meet the High threshold for biological and chemical capabilities.
  • Availability: Rolling out now in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, and in the API as gpt-6.2-astra and gpt-6.2-sol; Sol brings the Ultrafast option to Pro 200 subscribers for the first time.

Supply-Chain & Market Read-Through:

  • Cost curve: A frontier model delivering higher scores at lower prices directly cuts the unit cost of enterprise AI agents and long-horizon professional work, which may accelerate agent deployment in finance, law and design — and reshape inference purchasing, since lower cost per task can expand total task volume even as per-task compute falls.
  • Competitive pressure: Astra's lead over Claude Fable 5.2 on OpenAI's own benchmark, at 5% of the cost, puts direct pressure on Anthropic, which priced its IPO at $83 a share the same day and debuts on Nasdaq on 2026-11-19 09:30 ET / 14:30 UTC — model competition and capital-market competition converging on the same day.
  • Security side: The Critical cybersecurity rating plus tightened safeguards means higher-risk vulnerability research workflows will increasingly route through OpenAI's Daybreak defender programme, lifting demand for defender-side tooling and compliance.

Next Catalysts & Watchpoints:

  • 2026-11-19 09:30 ET / 14:30 UTC: Anthropic (ANT) begins trading on Nasdaq — watch how the market prices the two labs' competitive positions.
  • Independent benchmark updates from Artificial Analysis and Epoch AI to test whether OpenAI's self-reported lead holds on third-party evaluations.
  • The full GPT-6.2 system card from OpenAI, and progress on bringing GPT-6.2 models to the Daybreak programme.

Sources:

Information only, not investment advice.

→ 阅读中文版

More flash alerts from 2026-11-18

INSTANT ARCHIVE SEARCH

Search Stocks, Earnings & Macro Alerts

Quick tags: