OpenAI and Anthropic lowered mid-tier prices as customers test cheaper Chinese models, while DeepSeek announced higher V4-Pro rates from August 17. The contrast shows competition reshaping AI bills unevenly, by model tier and provider.
Artificial Intelligence··Midday
Mid-tier cuts, premium rates
OpenAI and Anthropic announced price reductions for mid-tier models. According to Ars Technica, OpenAI cut the price of GPT-5.6 Luna by 80%: the rate per million input tokens fell from 1 dollar to 0.20 dollars, and the rate per million output tokens fell from 6 dollars to 1.20 dollars. Anthropic introduced Claude Opus 5 at 5 dollars per million input tokens and 25 dollars per million output tokens, half the price of Fable 5. The company also cancelled an increase planned for Sonnet 5 in September. The Silicon Data token price index shows that prices paid for models from leading US labs have fallen by almost a quarter since mid-July. Mantas Lukauskas of Hostinger says prices for the best models are flat or rising. The reductions in the report concern GPT-5.6 Luna, Claude Opus 5 and Sonnet 5; a separate price trend is reported for the best models.[1]
DeepSeek raises rates with a new build
DeepSeek announced a higher tariff. According to Tech Wire Asia, the company released the V4-Pro-0813 build, described it as having improved agent capabilities, and announced a price list that takes effect on August 17. The announced increase ranges from 50% to 1,100% depending on the model, token type and time of day. DeepSeek is also introducing separate peak and off-peak rates; the tariff for the same model will vary with the time of use. On the earlier list, V4 Flash cost 0.14 dollars per million input tokens and 0.28 dollars per million output tokens. In independent measurements by Artificial Analysis, V4 Flash had beaten the earlier V4 Pro preview in several areas even though Pro was positioned as the more capable option. The report cites no independent measurement of whether the new V4-Pro-0813 build changes that picture. The source reports the new price list and the earlier independent measurement as separate information; it gives no independent performance result for the new build.[2]
Bills vary by provider and task
Ars Technica reports that OpenAI cut GPT-5.6 Luna by 80% and that Anthropic introduced Opus 5 at half the price of Fable 5. Tech Wire Asia reports that DeepSeek's V4-Pro-0813 will have increases from August 17 ranging from 50% to 1,100%. These tariffs belong to different providers and model tiers. Ars Technica also treats token price and cost per task as separate measures. The Artificial Analysis comparisons in the report match Opus 5 with Kimi K3 and GPT-5.6 Luna with DeepSeek V4 Flash on cost per task and performance. The effort level used in each match is also specified. DeepSeek's new list contains separate peak and off-peak rates. Provider, model tier, cost per task, effort level and time of use appear as separate information in the price comparison. The mid-tier reductions and the V4-Pro increase are reported as concurrent price movements for different products. Both sources tie the price changes to specific products.[1], [2]
