TL;DR
OpenAI and Anthropic launched competing model families within minutes of each other on September 22, 2026, signaling that the AI race has shifted from raw capability to cost-per-useful-task.
What happened
- Anthropic launched Claude Opus 5.5, the first model in its new 5.5 family, priced at $4 per million input tokens and $20 per million output tokens.
- OpenAI countered within minutes with GPT-6 Sol ($2 input / $10 output per million tokens) and GPT-6 Luna ($0.10 input / $0.50 output), both cut 50 percent versus GPT-5.6 promotional rates.
- Opus 5.5 costs 40 percent less to run on typical workloads than Opus 5 and generates output more than 30 percent faster.
- GPT-6 Sol and Luna extend the GPT-6 Astra architecture into cheaper, faster tiers, with context windows of roughly 1.05 million tokens and support for text and image inputs.
- Anthropic confirmed Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in coming weeks, mirroring OpenAI's tiered family strategy.
Why it matters
- Price is now the primary battlefield: OpenAI undercuts Anthropic at every comparable tier, with Sol costing half what Opus 5.5 charges per input token.
- Agentic and coding workloads are the target market: GPT-6 Sol scored 33.2 percent on AutomationBench at $0.27 per task, versus Claude Opus 5's 26.9 percent at more than 11 times the cost per task (OpenAI's figures).
- Factual accuracy is improving fast: GPT-6 Sol's internal coding-deception rate fell to 1.3 percent, down from 10.4 percent for GPT-5.6 Sol, a reduction that matters for enterprise deployments.
- Safety is a differentiator for Anthropic: Opus 5.5 underwent external testing by independent safety researchers and ships with added safeguards for cybersecurity and biology use cases, a posture designed to win regulated-industry buyers.
- Simultaneous launches compress decision cycles for enterprise buyers: procurement teams can no longer wait for a clear winner before committing to a platform.
What to watch next
- Whether Anthropic's Sonnet 5.5 and Haiku 5.5 close the per-token price gap with OpenAI's mid and low tiers when they ship in coming weeks.
- Real-world agentic benchmarks from neutral parties: both companies cite internal or vendor-produced evaluations (AutomationBench by Zapier, Agents' Last Exam), so independent replication will determine whether the cost-performance claims hold.
- Adoption curves in ChatGPT Free and Go tiers: Luna's availability to free users is a volume play that could lock in developer habits before Anthropic's cheaper models arrive.
Originally published on Present of AI, a daily source-linked AI news timeline. Read the full timeline or browse the open dataset.