OpenAI has quietly completed the GPT-6 family it started with Astra on September 3. Two new tiers now sit below it: GPT-6 Sol at $2.00/$10.00 per million tokens, and GPT-6 Luna at $0.10/$0.50. Both are confirmed directly against OpenAI's official pricing page, with no introductory discount or expiration listed for either. The headline result: GPT-6 Luna is now the cheapest model we track, full stop, ahead of even DeepSeek.
01The Full GPT-6 Rate Card
| Model | Input | Output | Cache read | Cache write | Batch in | Batch out |
|---|---|---|---|---|---|---|
| GPT-6 Astra | $10.00 | $50.00 | $1.00 | $12.50 | $5.00 | $25.00 |
| GPT-6 Sol | $2.00 | $10.00 | $0.20 | $2.50 | $1.00 | $5.00 |
| GPT-6 Luna | $0.10 | $0.50 | $0.01 | $0.125 | $0.05 | $0.25 |
All figures per 1 million tokens. Cache-write for Sol and Luna is shown as updated Sep 25, 2026 — it wasn't published as a separate line item at launch.
The tiering is unusually mechanical: each step down is exactly one-fifth of the tier above it, across every column, not just the headline input/output rate. Sol is 1/5th of Astra on input, output, cache-read, and both batch rates. Luna is 1/5th of Sol on all the same columns. That kind of clean, consistent ratio across an entire rate card is rare — most providers' tiers don't line up that precisely once you look past the headline number.
See what Sol or Luna costs for your actual usage
Model your real token volumes against every current model02GPT-6 Luna Beats DeepSeek Outright
DeepSeek V4.1-Flash has been the cheapest model on the market for most of September, at $0.15/$0.60 per million tokens off-peak. GPT-6 Luna changes that: $0.10/$0.50 is lower on both input and output, no off-peak caveat required.
| Model | Input | Output | Batch API |
|---|---|---|---|
| GPT-6 Luna | $0.10 | $0.50 | Yes — $0.05/$0.25 |
| DeepSeek V4.1-Flash (off-peak) | $0.15 | $0.60 | None published |
DeepSeek's rate shown is off-peak; peak-hour rates roughly double, which widens the gap further during those windows.
The gap isn't just the sticker price. DeepSeek has never published a Batch API for this tier, so its low base rate is effectively the ceiling for async workloads too. Luna has one: another 50% off, landing at $0.05/$0.25. For high-volume, batch-eligible work, that's a lower price on top of an already-lower price, not a marginal win. DeepSeek remains a reasonable choice for interactive, low-latency work, especially outside its peak UTC hours, but it's no longer the flat cheapest number to point to.
03GPT-6 Sol Slots in Against Gemini 3.1 Pro
Sol lands almost exactly where Gemini 3.1 Pro sits: $2.00 input matches Gemini's $2.00 exactly, while Sol undercuts Gemini on output, $10.00 versus $12.00. It's also a full 2x cheaper than GPT-5.6 Sol's own promotional rate ($4/$20) on input, and 2x on output too — a lower-numbered generation currently pricing below its own predecessor's discount rate, which is an unusual position for OpenAI to be in mid-lineup.
Whether Sol is a genuine substitute for GPT-5.6 Sol or Gemini 3.1 Pro on capability is a separate question this article isn't set up to answer — OpenAI's launch materials for GPT-6 Astra leaned on a benchmark methodology that didn't hold up to scrutiny (see our Astra pricing piece), so the same caution applies here: verify against your own task before assuming the price gap tracks a capability gap.
Compare Sol and Luna against every other model
Run your real token counts across GPT-6, GPT-5.6, Claude, Gemini, and DeepSeek04Update, Sep 25: Cache-Write Pricing Is Now Published
At launch, cache-write pricing wasn't broken out as a separate line item for either Sol or Luna — only cache-read was listed. We left it blank in our own data rather than guess, and flagged it for a recheck. OpenAI has since published it: $2.50 per million tokens for Sol, $0.125 for Luna — both confirmed directly against OpenAI's official pricing page and added to the table above and the live calculator. Both models also carry a Fast mode multiplier (2x standard on input/output: $4.00/$20.00 for Sol, $0.20/$1.00 for Luna), matching the pattern used elsewhere in the GPT-6 family.
05Where This Leaves the Rest of the Lineup
OpenAI now runs a three-tier GPT-6 family (Luna → Sol → Astra) alongside its existing GPT-5.6 lineup (Luna, Terra, Sol). That's six named tiers across two generations, several of which now overlap in price with each other or with competitors. For budget, high-volume work: GPT-6 Luna, now the outright cheapest option we track. For flagship-adjacent work at a lower price than the old flagships: GPT-6 Sol, priced against Gemini 3.1 Pro. For frontier work: Astra remains the top of the GPT-6 stack, unaffected by this launch.
None of this replaces running your actual token volumes through a calculator — the ranking above assumes clean input/output splits with no caching, which real workloads rarely match exactly.
Run your real numbers across every current model
Including caching, batch, and reasoning token overhead