TL;DR
OpenAI and Anthropic launched competing flagship-tier models within minutes of each other on September 22, 2026, with OpenAI cutting API prices 50% and claiming GPT-6 Sol beats Claude Opus 5.5 on every major agentic benchmark at a fraction of the cost.
What happened
- GPT-6 Sol and GPT-6 Luna went live minutes after Anthropic released Claude Opus 5.5, in what reads as a deliberate counter-launch.
- OpenAI cut API prices 50% versus GPT-5.6 promotional rates: Sol is now $2 per million input / $10 output; Luna drops to $0.10 input / $0.50 output.
- Sol and Luna sit below GPT-6 Astra (launched September 3) in OpenAI's lineup, mirroring Anthropic's Fable/Opus/Sonnet/Haiku tier structure.
- On AutomationBench (Zapier's 47-tool business workflow test), GPT-6 Sol scored 33.2% at $0.27 per task versus Claude Opus 5's 26.9% at more than 11 times the cost per task.
- On OSWorld 2.0 computer-use tasks, Sol scored 60.5% versus Claude Opus 5's 60.3%, at roughly 80% lower cost.
Why it matters
- Cost-per-task is now the battleground: raw benchmark scores are converging, so the team that delivers equivalent accuracy cheaper wins enterprise workloads at scale.
- OpenAI's coding-deception rate fell to 1.3% for Sol and 2.8% for Luna, down from 10.4% for GPT-5.6 Sol, a safety signal that matters for autonomous agent deployments.
- The simultaneous launch signals both labs are watching each other's release calendars and willing to compress timelines to deny the other a news cycle.
- Google is also in the fight: Gemini 3.7 Flash launched the same week at $0.75 input / $3.75 output (introductory through December 31), claiming AutomationBench scores above both Claude Sonnet 5 and GPT-5.6 Terra.
- Enterprise buyers now face a three-way price war across the mid-tier, with Google's introductory rate expiring January 1, 2027, creating a real procurement deadline.
What to watch next
- Whether independent benchmarks replicate OpenAI's cost-per-task claims for Sol versus Claude Opus 5.5, since all numbers cited come from OpenAI's own comparisons.
- Google's Gemini 3.5 Pro remains undelivered after four missed targets; if it ships before year-end it resets the flagship tier comparison entirely.
- How quickly enterprise contracts shift to Sol or Luna given the 50% price cut, which will show up in OpenAI's API revenue mix within one to two quarters.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.