Skip to main content

GPT-5.6 Launches with Three Tiers, Pricing War Heats Up AI Market

OpenAI today fully opened the GPT-5.6 series to the global market, launching three models at once: the flagship Sol, the balanced Terra, and the cost-effective Luna. The models will roll out across ChatGPT, Codex, and API within 24 hours. Pricing per million tokens (input/output) is $5/$30 for Sol, $2.5/$15 for Terra, and $1/$6 for Luna. The cheapest tier, Luna, costs just 1/16th of Anthropic's flagship Fable5.

Image

Performance that beats Claude, with coding and cybersecurity as key battlegrounds

The core message of GPT-5.6 is cost-effectiveness. In the Agents' Last Exam across 55 industries, Sol scored 53.6, beating Claude Fable5 by 13.1 percentage points. Even with medium reasoning, it outperformed by 11.4 points, at about a quarter of the cost. More surprisingly, Terra and Luna, positioned lower, still exceeded Fable5 at a fraction of the price. Programming is Sol's strong suit: it set a record of 80 points on the Artificial Analysis coding intelligence index, outputting less than half the tokens of Fable5 while cutting time and cost significantly.

But GPT-5.6 isn't flawless. On SWE-Bench Pro, Sol scored 64.6%, still below Fable5's 80% and Mythos5's 80.3%. On the toughest FrontierMath Tier4 problems, Sol's 65.9% even fell short of the previous GPT-5.5's 72.5%, while Fable5 hit 87.8%. These gaps show Anthropic still leads in deep engineering and frontier math. GPT-5.6 also introduces an ultra mode that coordinates up to 16 agents in parallel, setting a new record of 92.2% on BrowseComp, and its ExploitBench score in cybersecurity jumped from 47.9% to 73.5%.

China's GLM-5.2 emerges, creating a three-way rivalry

Chinese players aren't sitting idle. Zhipu AI's open-source GLM-5.2, released in mid-June, has reached international levels in coding and long-term tasks, supporting a million-token context. According to The Verge, several researchers say GLM-5.2's performance in vulnerability detection and cybersecurity testing is comparable to Anthropic's Mythos, showing Chinese AI models are rapidly closing the gap with the US in specialized fields. When Fable5 was shut down globally due to US export controls, GLM-5.2's stable open-source strategy offered developers a reliable alternative, making the three-way rivalry even more interesting.

Key Points

  • OpenAI's GPT-5.6 series includes three tiers: Sol (flagship), Terra (balanced), and Luna (budget).
  • Sol outperforms Anthropic's Fable5 in many benchmarks but lags in deep engineering and math.
  • Luna costs just 1/16th of Fable5, offering extreme cost-effectiveness.
  • China's Zhipu AI's GLM-5.2 emerges as a strong competitor, especially in cybersecurity.
  • The AI market is now a three-way battle between OpenAI, Anthropic, and Chinese players.