openai is about to join the price war started by deepseek

sam altman's lab is considering lower prices to win users from anthropic, per wsj – expecting anthropic to cut first

anthropic recently passed openai in business adoption for the first time: 34.4% vs 32.3% (ramp ai index)

chinese labs have been competing hard on price for months. xiaomi cut its mimo prices by up to 99% in may to match deepseek's level. now the us market looks ready for the same

what the wsj reported

the wall street journal reported that openai was weighing steep cuts to the price of its ai tokens. the talks were early, and no decision had been made.

openai expected anthropic to cut prices as well. both companies were preparing for an ipo, and business customers were pushing back on what ai costs them.

the pressure behind it

ramp's may index showed the first shift. in april, 34.4% of businesses on ramp paid anthropic and 32.3% paid openai. overall, 50.6% of them paid for some ai tool. the index counts companies that pay a vendor, not revenue.

price was the second pressure. the table above, from artificial analysis, shows the gap in june. claude fable 5 cost $8.20 per million tokens at a blended rate, gpt-5.5 cost $4.35, and deepseek v4 pro and xiaomi mimo-v2.5-pro cost $0.18.

openai sat in the middle. its flagship cost about half of anthropic's, but about 24 times as much as deepseek's.

update, october 16, 2026: who cut prices and who raised them

four months later, the price war split the market. the chart compares each lab's model line on june 11 and today: the same model, or its direct successor in the same tier. prices use the same 7:2:1 blended rate as the table above.

openai made the biggest cut. gpt-6.1 sol, the successor to gpt-5.5, costs $1.47, 66% less. at the same time openai added a new top tier above it: gpt-6 astra costs $7.70, 77% more than gpt-5.5 did.

anthropic got cheaper on both lines. opus 5.5 costs 24% less than opus 5. fable 5.1 kept fable 5's list price of $10 / $50, but cached input now costs $0.25 per million tokens, so the blended price fell 12%. mistral cut medium 3.5 by 45% without a new model. alibaba cut 18%, google 16% and nvidia 7%.

the cheap end moved up. deepseek v4 pro now costs $0.69 on deepseek's own api, 3.9 times the april version. kimi k3 costs 3.3 times as much as kimi k2.6, and grok 4.7 costs 2.1 times as much as grok 4.3. minimax, xiaomi, z.ai and amazon kept their prices.

the gap between the top and the bottom shrank. in june, gpt-5.5 cost 24 times as much as deepseek v4 pro. today gpt-6.1 sol costs about twice as much.

openai's cuts came in two steps. on july 30 it cut gpt-5.6 luna by 80% and gpt-5.6 terra by 20%. on september 22 it launched gpt-6 sol and luna at half the price of the 5.6 versions or less. the table shows list prices per million input and output tokens:

anthropic cut list prices only on opus: opus 5.5, released on september 22, is 20% cheaper per token than opus 5. fable's list price stayed at $10 / $50.

the cuts did not flip the adoption chart. in september, 43.8% of businesses on ramp paid anthropic and 40.6% paid openai. openai's share grew faster that month and narrowed the gap.

the wsj expected anthropic to cut first. openai moved first, almost two months earlier, and cut deeper. it also kept a premium tier for customers who pay for the top model.

sources

• openai mulls slashing prices ahead of competition from anthropic – cnbc on the wsj report, june 11, 2026
• ramp ai index, may 2026 – anthropic 34.4% vs openai 32.3%
• openai cuts prices for two of its gpt-5.6 models – cnbc, july 30, 2026
• openai api pricing – gpt-6.1 sol and gpt-6 luna
• artificial analysis llm leaderboard – list prices, june 11 and october 16, 2026
• ramp ai index, october 2026 – anthropic 43.8% vs openai 40.6% in september