The AI price war just went from quarterly salvos to same-day firefights, and your API bill is the casualty.
The Summary
- OpenAI dropped GPT-6 Sol and Luna with 50% lower API costs than GPT-5.6, released minutes after Anthropic shipped Claude Opus 5.5
- Perplexity immediately made GPT-6 Sol its default light option, signaling enterprise confidence in the new models
- The timing reveals AI labs now compete in real-time, not product cycles, compressing innovation windows to hours
The Signal
OpenAI launched two mid-tier models, Sol and Luna, at half the API cost of GPT-5.6 on the same day Anthropic released Claude Opus 5.5 at a lower price point than previous flagships. The coordination wasn't coincidence. These companies watch each other's product calendars like commodities traders watch Fed announcements. When Anthropic moved to undercut on price, OpenAI had Sol and Luna ready within minutes.
This is the new rhythm of AI competition. Not quarterly earnings calls or annual developer conferences, but same-day tactical responses. The gap between "we're working on something" and "it's live in production" has collapsed to the point where launch timing is now a competitive weapon.
"AI rivalry is now measured in minutes, not months."
Perplexity's decision to default to GPT-6 Sol for light tasks tells you something the press releases won't: the model works. Consumer AI companies don't swap defaults on launch day unless they've been testing in private for weeks. Perplexity's move suggests:
- Sol's performance meets or exceeds GPT-5.6 despite the lower price
- The cost savings are material enough to change unit economics
- OpenAI pre-briefed key partners to coordinate the rollout
The 50% price cut hits different than previous reductions. Earlier drops came as models got more efficient. This one arrives while OpenAI is still burning capital to train frontier models. The implication: margin compression is a strategy, not a side effect. They're trading profitability for market position, betting that cheaper inference creates stickier customers.
The Implication
If you're building on foundation models, your cost assumptions just changed. The 50% reduction in API costs makes agent swarms, real-time analysis, and always-on AI assistants viable at scales that were underwater last week. Rebuild your spreadsheets. Products that didn't pencil out at GPT-5.6 pricing might now.
Watch what happens when Anthropic responds. If they cut again, we're in a race to the bottom that ends with inference as a loss leader and the real money in fine-tuning, enterprise contracts, and proprietary data moats. The companies that win won't be the ones with the cheapest API. They'll be the ones who figured out what to build when compute costs stop mattering.