OpenAI Cut Luna's API Price 80%. DeepSeek Is Still Cheaper.
OpenAI cut GPT-5.6 Luna's API price 80% on July 30, three weeks after setting it. DeepSeek V4-Flash still beats the new rate on both input and output.
Published: August 1, 2026 Impact: High. Affects developers and businesses paying for ChatGPT API access by the token, and anyone comparing model costs across vendors.
A Price OpenAI Set Itself
On July 30, 2026, OpenAI cut GPT-5.6 Luna's API pricing 80%: from $1.00 input / $6.00 output to $0.20 input / $1.20 output per million tokens. OpenAI set the higher price itself three weeks earlier, on July 9, when GPT-5.6 launched.
OpenAI's 80% cut is measured against a price OpenAI itself set three weeks earlier, and even after the cut, DeepSeek's cheapest model still beats it on both input and output.
Most coverage in the last two days ran the cut as a generosity story. It's a baseline story. And the new price still isn't the market's cheapest.
The New Numbers
Three GPT-5.6 tiers, three different moves on July 30:
- Luna (cheapest tier): $1.00/$6.00 → $0.20/$1.20 (per million input/output tokens), an 80% cut on both ends
- Terra (mid tier): $2.50/$15.00 → $2.00/$12.00, a 20% cut
- Sol (top reasoning tier): $5.00/$30.00, unchanged
Sol also gained a new Fast Mode: roughly 2.5x the response speed at 2x the price, replacing what OpenAI previously called Priority Processing.
Still Not the Cheapest
I checked DeepSeek's current API pricing directly against its own docs on August 1. Here's the comparison that matters:
| Model | Input ($/M tokens) | Output ($/M tokens) |
|---|---|---|
| New GPT-5.6 Luna (Jul 30) | $0.20 | $1.20 |
| DeepSeek V4-Flash | $0.14 | $0.28 |
| GPT-5.4 Nano | $0.20 | $1.25 |
New Luna's input price matches GPT-5.4 Nano to the cent. Its output price sits 4% below Nano, essentially the same tier repriced. Against DeepSeek V4-Flash, new Luna still loses on input (43% higher) and loses badly on output (over 4x higher). The 80% cut closed the gap with OpenAI's own prior generation. It didn't close the gap with the market's cheapest model.
OneHuman's Pro Deep Analysis tracks this comparison across all 8 tools and 4 cost dimensions, updated every 15 days.
What OpenAI Says
OpenAI frames the cut as an efficiency gain: improved serving infrastructure lets it pass savings to developers. That's plausible, and if true, a real benefit to anyone paying by the token. The part worth naming is what OpenAI measured the cut against: its own three-week-old price, not the competitive field. Measured against the field, Luna caught up to itself. It did not take the lead.
Consumer Protection Q&A
Q: Is 80% cheaper actually cheap? A: Cheaper than three weeks ago, yes. Cheaper than DeepSeek V4-Flash, no, on either input or output.
Q: Did OpenAI lower prices because of competition? A: OpenAI hasn't said so, and I'm not asserting it did. The checkable fact is where the new price lands relative to the field, not why OpenAI moved.
Q: Does this change what Sol costs? A: No. Sol's per-token price is unchanged. Its new Fast Mode costs 2x the standard rate for roughly 2.5x the speed.
What to Do Next
- Building on the API, cost-sensitive: Compare Luna, Nano, and DeepSeek V4-Flash against your actual token mix before assuming the July 30 cut makes Luna the cheap option by default.
- Using Sol for reasoning workloads: Nothing changed on price. Fast Mode is a speed-for-cost trade, not a discount.
- Comparing vendors: OneHuman's comparison table tracks per-token pricing across all 8 tools; Pro Deep Analysis breaks it down further by cost dimension.
Sources
- OpenAI, "Advancing the price-performance frontier with GPT-5.6" (July 30, 2026). Direct automated access to openai.com returned 403 on verification attempts July 30 and 31; figures confirmed via two independent OneHuman news-monitor briefs citing the same post, and via a prior independent multi-source check (OpenAI's blog, Axios, CNBC, VentureBeat, Quartz) completed July 31.
- DeepSeek V4-Flash pricing: verified directly against api-docs.deepseek.com, August 1, 2026.
- GPT-5.4 Nano pricing: third-party pricing tracker, independently checked July 31, 2026.
- OneHuman's own compare-tools data, current as of August 1, 2026.
Verified by OneHuman · August 1, 2026
Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.
Share This Article
"OpenAI's 80% price cut on GPT-5.6 Luna is measured against a price OpenAI set itself three weeks earlier. That's not generosity math. That's a repricing back into the market band."
"New Luna API pricing: $0.20 input, $1.20 output per million tokens. DeepSeek V4-Flash: $0.14 and $0.28. Even after an 80% cut, OpenAI's cheapest model still loses on both numbers."
"GPT-5.6 Luna's new price matches GPT-5.4 Nano on input to the cent and lands 4% above it on output. OpenAI didn't invent a new price floor. It found the one that already existed."
"Sol held flat at $5.00 input, $30.00 output while Luna and Terra got cut. OpenAI is protecting its frontier margin and conceding the commodity floor to prices the market already set."
Related: ChatGPT
- • ChatGPT, Codex, and Atlas Just Became One App. Here's What You're Agreeing To.
- • OpenAI Just Built an Ad Platform Inside ChatGPT — Here's What That Changes for the 800 Million People Already Using It
- • OpenAI Puts Codex on Your Phone — Free. Anthropic Wants $20/Month.
- • ChatGPT Got Five Upgrades in Four Days — Did You Notice?
- • GPT-5.5 vs. The Seven-Day War: API Prices Double, GitHub Collapses, and Why This Pace Can't Last