OpenAI prices GPT-6 Sol at $2 and $10 per million tokens. The 50% cut does not apply everywhere

OpenAI released two new models on Tuesday, GPT-6 Sol and GPT-6 Luna, and cut their API prices in half compared with the promotional prices on the GPT-5.6 versions they replace. Sol, the middle tier below the flagship GPT-6 Astra, now lists at $2 per million input tokens and $10 per million output tokens, down from $4 and $20. Luna, the budget tier for high-volume work such as extraction and summaries, drops to $0.10 and $0.50 from $0.20 and $1.20, according to OpenAI's announcement.

The launch came hours after Anthropic released Claude Opus 5.5. CNBC noted both labs face pressure from cheaper open-weight models from Chinese firms including Alibaba, Moonshot AI and DeepSeek, and Fortune quoted Ramp economist Ara Kharazian describing the two companies as locked in a price war.

What the price sheet shows

The $2 and $10 figures are the standard rate for short prompts. OpenAI's API pricing page lists separate columns, and two of them take the headline saving back:

• Fast mode costs exactly what the old model cost. Sol in fast mode (OpenAI renamed "priority processing" to fast mode in July) is $4 in and $20 out, the same as GPT-5.6 Sol. A team that pays for speed gets a newer model, not a cheaper one.
• Long prompts double the input price. Long-context input is $4, again matching the old short-context rate. Agents and coding tools that feed in whole codebases or long document sets are the users most likely to land here.
• Batch is where the real discount is. Jobs that can wait run at half the standard rate, $1 and $5.
• The old promotion is not over. The page says GPT-5.6 Sol's promotional pricing runs at least through November 21, 2026, so the comparison OpenAI is drawing is against a discount that is itself still live.

How it compares with Anthropic's launch the same day

At the standard short-context rate, Sol lists at half of Claude Opus 5.5, which Anthropic priced at $4 and $20 (our breakdown of that launch). Sol in fast mode costs the same as Opus 5.5 at its standard rate.

OpenAI's benchmark claims also need reading carefully. Its release says Sol beats Claude Opus 5 on a business-workflow test at 9% of Opus 5's cost per task, but the comparison is against Opus 5, not the Opus 5.5 that shipped the same morning. The competitor scores come from public reports, and OpenAI notes one Anthropic cost figure leaves out fallback costs. Those numbers are the company's own and have not been independently checked.

What it means for a business paying the bill

A company running customer-service replies or document summaries at normal speed should see close to the full 50% on Sol or Luna. A software team running coding agents on long contexts, or anyone paying for fast responses, should budget at roughly the old price. For a workload of 10 million input and 2 million output tokens a month, standard short-context Sol comes to $40, against $80 in fast mode, using only the list prices above.

For investors, the more useful signal is Kharazian's point in Fortune: prices at the two leading labs are falling, not rising, as models improve. That matters for any valuation built on AI revenue per customer growing over time, including Anthropic's planned listing.

Sources: OpenAI, OpenAI API pricing, CNBC, Fortune. Benchmark results are OpenAI's own. This is market information, not investment advice.