OpenAI cuts GPT-6 API prices by 50%; Box CEO sees expanded agent use
- OpenAI launched GPT-6 Sol and Luna with API prices cut by up to 50% versus GPT-5.6
- Box CEO Aaron Levie stated lower AI costs could dramatically expand agent use cases
- Anthropic's Claude Opus 5.5 is 40% cheaper than its predecessor, intensifying competition
- Survey data shows nearly 60% expect AI to handle more work tasks within a year

*this image is generated using AI for illustrative purposes only.
OpenAI introduced GPT-6 Sol and GPT-6 Luna on Tuesday, cutting API prices by up to 50% compared with GPT-5.6 promotional pricing. Box Inc. (NYSE: BOX) CEO Aaron Levie said the cost decline could significantly expand the number of tasks companies can economically automate with AI agents.
The announcements arrive hours after rival Anthropic unveiled Claude Opus 5.5, which costs 40% less to run than its predecessor while improving performance in coding and knowledge work. Both firms are prioritizing cost-per-task efficiency as enterprises shift from chatbots to long-running agentic workflows.
Pricing structure shifts
OpenAI positioned GPT-6 Astra as its highest-performance model, while Sol and Luna target workloads where speed and operating costs outweigh maximum benchmark scores. The company reported that GPT-6 Sol scored 33.2% on AutomationBench at its highest reasoning setting, costing 27 cents per task. In comparison, GPT-6 Astra scored 30.3% at low effort but cost 3.9 times as much per task.
| Model | Input Cost (per million tokens) | Output Cost (per million tokens) | Change vs Previous |
|---|---|---|---|
| GPT-6 Sol | $2 | $10 | Down from $4 / $20 |
| GPT-6 Luna | $0.10 | $0.50 | Down from $0.20 / $1.20 |
Industry reaction to cost declines
Levie described the simultaneous price reductions by OpenAI and Anthropic as "an insane day in AI" on social media platform X. He argued that the rate at which the cost per task drops in AI is unlike any other type of technology in history.
According to Levie, every time the cost of AI drops, the use cases available for agents increase dramatically. He described this trend as "Jevons paradox applied to agents," suggesting that lower costs lead to broader diffusion of AI in the economy. Specific applications he cited include processing data, scanning code for security issues, reading log data, and managing agent swarms in workflows.
Broader adoption trends
Recent industry data supports the shift toward agentic workflows. In June, Anthropic reported that close to six in 10 survey respondents expected AI to handle a greater share of their work tasks within a year, while more than one-third expected AI to be capable of performing most or nearly all of their work tasks.
Earlier this year, former Tesla Inc. (NASDAQ: TSLA) AI chief and ex-OpenAI researcher Andrej Karpathy said AI agents had dramatically changed his coding workflow, noting he had barely written code since December. He also built "Dobby," an AI assistant that managed smart-home functions, including lighting, climate control, security monitoring and delivery alerts.
Efficiency and error reduction
OpenAI stated that GPT-6 Sol made approximately half as many errors as its predecessor in an internal factuality test based on de-identified user conversations. The company also expanded prompt-caching tools to reduce costs for applications that repeatedly send identical context to the model.
Anthropic adopted a similar strategy, noting that cache reads constitute a majority of costs for agentic and coding workloads. Consequently, Anthropic reduced its cache-read price by 60% from Opus 5.
What the numbers show
The pricing data reveals a strategic bifurcation in model economics. While GPT-6 Sol offers a 50% input cost reduction compared to the previous tier, its output cost also halves, indicating a uniform compression across token types rather than a skew toward input-heavy tasks. Furthermore, the disclosed cost-per-task metric highlights a significant efficiency gap: GPT-6 Sol delivers a higher benchmark score (33.2% vs 30.3%) at roughly one-fourth the cost of GPT-6 Astra (27 cents vs implied ~$1.05 based on the 3.9x multiplier), suggesting that mid-tier models are becoming disproportionately more valuable for automated workflows.
How will the 50% API price reduction impact the unit economics of AI startups relying on high-volume agent workflows?
Will the aggressive pricing war between OpenAI and Anthropic force smaller model providers to exit the enterprise market or pivot to niche verticals?
What regulatory or security challenges might arise as lower costs enable the widespread deployment of autonomous 'agent swarms' in critical infrastructure?































