OpenAI and Anthropic Slash AI Prices as DeepSeek and Moonshot Gain Ground
OpenAI and Anthropic cut mid-tier model prices as Moonshot and DeepSeek gain ground, while usage billing drives DoorDash and Airbnb toward Chinese AI.
Summary
On August 14, 2026, OpenAI said it was cutting GPT-5.6 Luna, its fastest, cheapest model, by 80 percent, from $1 to $0.20 per million input tokens and $6 to $1.20 per million output tokens. Anthropic launched Claude Opus 5 at $5 and $25 respectively, half Fable 5's rates, and canceled a Sonnet 5 increase due in September. Silicon Data's token price index says leading US labs' customer prices have fallen almost 25 percent since mid-July. Tokens measure model inputs and outputs for billing.
Rising AI bills, amplified as OpenAI and Anthropic move some enterprises from flat subscriptions to usage-based charges for computing consumed, have prompted caps and cheaper trials. DoorDash and Airbnb now use Chinese models to control costs, while increasingly capable, downloadable and modifiable releases from Moonshot and DeepSeek narrow the gap and win users from Silicon Valley to Europe. The cuts target mid-tier products as proprietary US labs defend premium pricing, spend heavily on performance, and plan IPOs at trillion-dollar valuations under investor pressure to prove returns.
Token rates remain imperfect because model versions and effort settings alter compute, performance, attempts and total task cost. Artificial Analysis found Opus 5 at medium effort matched Moonshot's Kimi K3 at max effort on performance and per-task cost. GPT-5.6 Luna at max matched DeepSeek V4 Flash at max but cost just under twice as much. Mantas Lukauskas, AI tech lead at Hostinger, which has used large language models since 2020, said top-model prices remain flat or rising. Both labs declined comment, while a person close to Anthropic said Opus 5's discount reflects its model family, not competitors.
Positives
- GPT-5.6 Luna input and output prices fell 80 percent, to $0.20 and $1.20 per million tokens.
- Claude Opus 5 launched at half Fable 5's prices, charging $5 for input and $25 for output per million tokens.
- Anthropic canceled a planned Sonnet 5 price increase that was due to begin in September.
- Leading US labs' customer prices have dropped almost 25 percent since mid-July, according to Silicon Data.
- Opus 5 at medium effort matched Kimi K3 at maximum effort on performance and per-task cost, Artificial Analysis found.
Risks & concerns
- GPT-5.6 Luna matched DeepSeek V4 Flash at maximum effort but cost just under twice as much per task.
- DoorDash and Airbnb have adopted Chinese models as rising AI bills push businesses toward cheaper alternatives.
- OpenAI and Anthropic are moving some enterprise customers from flat subscriptions to potentially less predictable usage-based billing.
- Moonshot and DeepSeek are narrowing the performance gap while US labs continue spending heavily to preserve their technological advantage.
- Top-model prices remain flat or rising despite mid-tier cuts, according to Hostinger AI tech lead Mantas Lukauskas.