On September 22, 2026, Anthropic and OpenAI both released new models on the same day, as CNBC reported.
- This guide provides comprehensive, actionable information
- Consider your specific workflow needs when evaluating options
- Explore our curated LLMs tools for specific recommendations
What actually shipped
On September 22, 2026, Anthropic and OpenAI both released new models on the same day, as CNBC reported. Anthropic shipped Claude Opus 5.5, calling it a model that performs at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5. OpenAI shipped GPT-6 Sol and Luna, its coding-focused and high-volume clerical models, at roughly half the price of the GPT-5.6 versions they replace. Both companies framed the releases the same way: cheaper, not smarter.
The timing is the story. Ten days earlier, on September 12, 2026, Anthropic CEO Dario Amodei had publicly called for the industry to slow the pace of frontier capability gains, warning that unchecked progress risked producing agents capable of "taking over the entire internet" within six to twelve months, a position OpenAI's Sam Altman and xAI's Elon Musk both said they agreed with. These are the first model releases from either lab since that pledge, and neither company shipped a capability jump. Both shipped a price cut.
The actual numbers, not the framing
Anthropic's pricing: Opus 5.5 runs $4 input / $20 output per million tokens, down from $5 / $25 for Opus 5, a 20% cut on the headline rate. Cache reads dropped further, from $0.50 to $0.20 per million tokens, which Anthropic says is the bigger deal since cache reads dominate the cost of agentic and coding workloads — that's where the 40%-cheaper-on-typical-workloads claim comes from. A faster mode is also available in Claude Code at $8 input / $40 output, running up to 2.5x faster. On benchmarks, Anthropic's own figures show Opus 5.5 ahead of Fable 5.1 on agentic coding (66.4% vs 55.8% on Terminal-Bench 4.0) and computer use (81.8% vs 80.7% on OSWorld 2.0), while running at less than half Fable's price.
OpenAI's pricing: GPT-6 Sol dropped from $4/$20 to $2/$10 per million input/output tokens; GPT-6 Luna dropped from $0.20/$1.20 to $0.10/$0.50. An OpenAI spokesperson confirmed to VentureBeat these are permanent prices, not promotional. On the DeepSWE v1.1 coding benchmark, independent testing by The Decoder found GPT-6 Sol reaches 68.8% at maximum effort for $2.14 per task, against Fable 5's 69.9% at "xhigh" effort for $21.63 per task — essentially matching score at a tenth of the cost. The same outlet's broader read: "intelligence scores stay at GPT-5.6 levels, with gains in some evaluations and regressions in others."
Cheaper is not nothing, but it is not what the pledge was about
A 20-50% price cut on frontier-adjacent models is a real, usable change if you run agentic workloads at volume — it directly lowers the cost of the exact coding and computer-use tasks these models are priced around. But the slowdown pledge Amodei made on September 12 was about capability gains, not cost. Neither release moves the needle on what the models can do that Opus 5 or GPT-5.6 could not; both labs are explicit that these are cost and efficiency plays layered onto existing capability tiers. Whether that counts as compliance with a slowdown or simply the next scheduled price war is a question the releases themselves don't answer — there's no new capability benchmark being suppressed here, because there's no new capability being claimed.
Who should actually switch
For the full pricing landscape across every major LLM, see LLM pricing: complete cost comparison. For managing spend once you've picked a model, see AI tool cost optimization. For the tools themselves, see every LLM in the directory.
Explore curated tools related to this guide: