Ai Engineering 4 min read

OpenAI Cuts Flagship Prices in Half With GPT-6 Sol and Luna

OpenAI launched GPT-6 Sol and Luna on September 22 and halved flagship pricing: Sol drops to $2 per million input and $10 output, while Luna lands at $0.10/$0.50, the cheapest frontier-lab tier yet, days after Grok 4.7 opened a price war.

OpenAI launched GPT-6 Sol and GPT-6 Luna on September 22 and, per The New Stack and VentureBeat, cut its token prices roughly in half at the same time. Sol, the frontier tier, moves to $2 per million input and $10 output from GPT-5.6 Sol’s $4/$20. Luna, the high-volume tier, lands at $0.10/$0.50 per million, down from $0.20/$1.20. That is a 50% cut on the flagship and up to 58% on the budget model, announced two days after Grok 4.7 launched at $2/$6 declaring a price war and hours after Anthropic shipped Opus 5.5 at $4/$20 with the week’s best benchmarks. Three frontier labs repriced the entire market in seventy-two hours.

Reading the New Price Sheet

The Sol cut is defensive and the Luna cut is offensive, and it is worth separating them. Sol at $2/$10 matches Grok 4.7’s input price while conceding output (xAI charges $6), which reads as OpenAI refusing to be the expensive option on anyone’s comparison chart. Note what OpenAI did not do: it did not try to beat Grok’s headline numbers outright, and it did not discount GPT-6 Astra, the capability leader from earlier this month, which still commands $10/$50. The flagship lineup is now stratified by price rather than by generation: Astra holds the premium, Sol takes the volume battleground, and the release’s timing (between Grok’s launch and Opus 5.5’s benchmark coronation) suggests the pricing was the message.

Luna’s cut is the one with longer consequences. At $0.10 input and $0.50 output per million tokens, the cheapest tier from a frontier lab now costs less than most open-weights models cost to self-host once you count engineering time. Concretely: a million-token agent session (long context, heavy tool output, multiple revisions) costs fifty cents at Luna rates versus twenty dollars at Fable 5.1-class pricing. High-volume workloads that were economics experiments last month, classifier fleets, always-on monitoring agents, cache-heavy document pipelines, become line items nobody audits.

What This Does to the Market Structure

Three pricing regimes now coexist at the frontier, and each lab has picked a different answer to the same question. Anthropic says capability still carries a premium and points at benchmark crowns (Opus 5.5 leads most boards this week). xAI says price is the product and points at Grok’s $6 output. OpenAI is running both plays at once: Astra defends the top, Sol and Luna take the mid and bottom. For buyers this resolves into a routing problem rather than a loyalty problem, and the winners are the platforms already built to route: coding tools, agent harnesses, and API aggregators can now pick per-task between a $0.50 and a $50 output tier with genuinely frontier options at both ends.

The open-weights side should pay attention too. MiMo v2.6’s MIT-licensed debut looked unbeatable on price at $0.435/$0.87 per million on Monday; Luna’s $0.10/$0.50 with zero self-hosting overhead erases most of that gap without ever publishing a weight.

What to Watch

First, independent benchmarks for Sol specifically: the launch coverage leads with price, and until third parties publish Sol versus Opus 5.5 and Grok 4.7 numbers, nobody knows whether the $2 tier is capability-matched or a repositioned mid-model. Second, margin disclosure: three simultaneous price cuts at the frontier, on top of this summer’s discounts, imply inference costs falling faster than most financial models assumed, which will show up in the labs’ next revenue reports. Third, whether Anthropic’s next move is a Haiku 5.5 price point below Luna’s, because the one tier nobody has attacked yet is the sub-dollar one OpenAI just claimed. The week’s lesson is already clear though: capability differentiation at the frontier now lasts days, and price is where the competition has moved to stay.

Get Insanely Good at AI

Get Insanely Good at AI

The book for developers who want to understand how AI actually works. LLMs, prompt engineering, RAG, AI agents, and production systems.

Keep Reading

Ai Engineering

What Are Parameters in AI Models?

Parameters are the numbers that make AI models work. Here's what they are, why models have billions of them, and what the count actually tells you about capability.

Ai Engineering

GPT-6.1 Sol Arrives With Near-Astra Performance at a Fifth of the Price

Announced at DevDay on September 29, GPT-6.1 Sol claims near-GPT-6 Astra performance at $2/$10 per million tokens, with a 1.05 million token context window and a new Ultrafast tier at 300 tokens per second, one week after GPT-6 Sol launched.

Ai Engineering

Gemini 4 Hydrogen replaces Flash at $1.50/$7.50: faster agents, triple the Terminal-Bench score, and a higher bill per token

Google released Gemini 4 Hydrogen on November 12 as its everyday model for coding and agents. It scores 51 on the Artificial Analysis index, beats Argon, costs twice Flash's current price, and its model card admits more unjustified refusals.

Ai Engineering

Gemini 4 Carbon reaches the API: $2/$10 until November 30, then double, and a safety report that calls it no material change

Google opened Gemini 4 Carbon to paying developers and AI Ultra subscribers on November 10, with a 1M-token window and a price rise to $4/$20 on November 30. Its safety report carries Argon's risk findings forward, while showing a 12.8-point jump in exploit success.

Ai Engineering

Meta's Muse Ember matches Haiku 5.5 and GPT-6 Luna on list price and competes on tokens per task

Meta released Muse Ember on November 9 at $0.10 per million input tokens and $0.50 output, the same rates as Claude Haiku 5.5 and GPT-6 Luna. Its pitch is cost per task: about $0.14 on the Artificial Analysis index against $0.41 for Haiku at a similar score.