# OpenAI Cuts GPT-5.6 Sol API Prices After Holding the Line for Months > Source: > Published: 2026-08-22 01:35:01+00:00 *OpenAI just cut prices on its flagship GPT-5.6 Sol model for developers, the first price cut to its top-tier model since launch and a clear sign the AI price war is squeezing even the company that helped start it.* OpenAI announced on Friday, August 21, that it's cutting API pricing for GPT-5.6 Sol by more than 20% for the next three months. Input tokens drop from $5 to $4 per million. Output tokens fall from $30 to $20 per million. The cheaper rates apply to the API, to credits on the agentic ChatGPT Work product, and to Codex, OpenAI's coding tool, according to a report from Reuters. Subscription pricing for Plus, Pro, and Business plans doesn't change. Sol had held out. When OpenAI cut prices on its mid-tier Terra model by 20% and its budget Luna model by 80% on July 30, Sol stayed put at its launch rate, according to CNBC. That made this week's move the second pricing cut to the GPT-5.6 family since the models launched on July 9, and the first time OpenAI has touched the price of its best model rather than just the cheaper ones underneath it. Here's why that matters. Cutting Luna and Terra prices is a rounding error for OpenAI's business. Sol is the model enterprise customers actually pay real money to run at scale, and it's the one OpenAI has spent two years positioning as worth a premium over everyone else's frontier offering. Discounting it, even for three months, is an admission that the premium isn't holding. The pressure is coming from two directions at once. Anthropic prices its Claude Sonnet 5 at an introductory $2 per million input tokens and $10 per million output, rising to $3 and $15 after August 31. Its frontier Claude Opus 5 runs $5 and $25. Those numbers sit below where Sol was pricing just last week, and Anthropic has been converting that gap into enterprise API share. [The EU AI Act Now Forces Every Chatbot to Admit It Is Not Human](https://startupfortune.com/the-eu-ai-act-now-forces-every-chatbot-to-admit-it-is-not-human/) The EU's AI Act transparency rules became enforceable on August 2, 2026, and regulators began active enforcement this week, forcing every chatbot, AI agent and deepfake reaching EU users to disclose it isn't human. Fines reach €15 million or 3% of global turnover, and while 180-plus firms including Anthropic, OpenAI and Google signed onto a... - [EU AI Act chatbot disclosure requirements](https://startupfortune.com/the-eu-ai-act-now-forces-every-chatbot-to-admit-it-is-not-human/) - [AI transparency rules customer service compliance](https://startupfortune.com/the-eu-ai-act-now-forces-every-chatbot-to-admit-it-is-not-human/) Then there's China. DeepSeek V4 Pro delivers a million output tokens for about $0.87. Moonshot AI's Kimi K3 charges $15 for the same volume, according to reporting compiled by BigGo Finance. DoorDash has already moved coding workloads onto Moonshot AI's models. Siemens and Airbnb are testing DeepSeek and Alibaba's offerings for at least part of their workloads. None of these companies are switching because they love the branding. They're switching because the invoices are smaller. OpenAI knows this. Its own July blog post framed the Terra and Luna cuts as efficiency gains, crediting better hardware routing, improved inference software, and smarter context caching, and claiming Sol had been rewriting its own production code kernels to trim serving costs by roughly 20%. That's a real technical story. It's also, conveniently, the exact justification a company needs when it's about to cut prices under competitive pressure rather than out of pure generosity. Frankly, calling this three-month window a temporary promotion rather than a permanent cut is doing a lot of work. It lets OpenAI claim flexibility while it watches how much developer traffic actually returns. If usage climbs and stays up once the window closes in November, expect the lower Sol pricing to quietly become the new normal. If it doesn't move the needle, OpenAI has an easy off-ramp back to $5 and $30. What this confirms is something the market has suspected since Anthropic's enterprise API growth started showing up in analyst notes and DeepSeek's V4 Pro pricing started circulating on cost comparison spreadsheets: frontier model margins bend once the pressure gets heavy enough, and OpenAI is the one bending them now. Compute costs for running Sol haven't dropped 20% in three weeks. What's changed is who else developers can call instead. Three months from now, OpenAI will decide whether $4 and $20 per million tokens is where Sol lives permanently, or whether this was a defensive flare sent up while Anthropic and the Chinese labs were watching. Either way, the era of OpenAI setting the price and letting everyone else chase it looks over. **Also read:** [Nvidia Shows a Better AI Harness Beat a Smarter Model on ARC-AGI-3](https://startupfortune.com/nvidia-shows-a-better-ai-harness-beat-a-smarter-model-on-arc-agi-3/) • [Dutch Regulator Fines Uber Nearly $1 Billion Over Automated Driver Suspensions](https://startupfortune.com/dutch-regulator-fines-uber-nearly-1-billion-over-automated-driver-suspensions/) • [Reddit May Cut Off Google's AI Access When Their $60 Million Deal Expires](https://startupfortune.com/reddit-may-cut-off-googles-ai-access-when-their-60-million-deal-expires/)