--- title: "GPT-5.6 promo did not expire: Luna stays $0.20/$1.20 and Sol $4/$20" description: "Correction: GPT-5.6 promo remains active. Sol is $4/$20, Terra $2/$12, and Luna $0.20/$1.20 per MTok. Prices verified 2026-09-13." canonical: https://vynaris.com/blog/gpt-56-promo-expired-sol-5-30-terra-2-5-15-luna-1-6 author: "Dhrumil Barot" author_bio: "Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers." published: 2026-09-11 updated: 2026-09-13 --- # GPT-5.6 promo did not expire: Luna stays $0.20/$1.20 and Sol $4/$20 Published 2026-09-11 · Updated 2026-09-13 · By [Dhrumil Barot](https://vynaris.com/about#founder), Founder Correction: GPT-5.6 promo remains active. Sol is $4/$20, Terra $2/$12, and Luna $0.20/$1.20 per MTok. Prices verified 2026-09-13. *Updated 2026-09-13: This article previously said the GPT-5.6 promotion had expired. That was wrong.* [GPT-5.6 Sol](https://vynaris.com/models#gpt-5-6-sol) remains $4 input and $20 output per million tokens. [GPT-5.6 Terra](https://vynaris.com/models#gpt-5-6-terra) remains $2/$12, while [GPT-5.6 Luna](https://vynaris.com/models#gpt-5-6-luna) remains $0.20/$1.20. Prices verified 2026-09-13. ## TL;DR - OpenAI's live pricing page says Sol promotional pricing remains available at least through 2026-11-21. - A 10k-input/2k-output Standard call costs $0.0800 on Sol, $0.0440 on Terra, and $0.0044 on Luna. - Those figures are 27.3%, 20.0%, and 80.0% below the incorrect costs implied by our prior rate card. - We have replaced the article in place. This is a correction, not a newly announced price cut. ## Corrected verdict table | Model | Current input/output per MTok | Incorrect rate previously shown | Current 10k/2k task | Incorrect 10k/2k task | Correction | | --- | --- | --- | --- | --- | --- | | gpt-5.6-sol | $4 / $20 | $5 / $30 | $0.0800 | $0.1100 | 27.3% lower | | gpt-5.6-terra | $2 / $12 | $2.5 / $15 | $0.0440 | $0.0550 | 20.0% lower | | gpt-5.6-luna | $0.20 / $1.20 | $1 / $6 | $0.0044 | $0.0220 | 80.0% lower | The workload uses 10,000 uncached [input tokens](https://vynaris.com/glossary/category/cost-and-finops#input-token-cost) and 2,000 [output tokens](https://vynaris.com/glossary/category/cost-and-finops#output-token-cost) per call. For Sol, `(10,000 × $4 + 2,000 × $20) / 1,000,000 = $0.08`. The correction is `($0.11 - $0.08) / $0.11 = 27.3%`. These are editable assumptions, not observed Vynaris traffic. Source: OpenAI API pricing and the previous version of this article. Prices verified 2026-09-13. ![Current GPT-5.6 cost per 1,000 tasks](https://cdn.sanity.io/images/u1qdn8dx/production/a870b15e440ff2ca04668cb92ed43c2e652a5cbc-1177x734.png?w=1440&fit=max&auto=format) *Source: OpenAI API pricing, verified 2026-09-13. 8k input / 2k output per task. Log scale.* ## What the live page says OpenAI's [official API pricing page](https://developers.openai.com/api/docs/pricing) lists the current Standard short-context rates. Sol is $4 input, $0.40 cached input, $5 cache write, and $20 output per MTok. Terra is $2/$0.20/$2.50/$12. Luna is $0.20/$0.02/$0.25/$1.20. The same page states that Sol's promotional pricing is available at least through 2026-11-21. The model IDs shown there are `gpt-5.6-sol`, `gpt-5.6-terra`, and `gpt-5.6-luna`. That checks both the prices and the names against one primary source. OpenAI's [2026-07-30 announcement](https://x.com/OpenAI/status/2082878156483219672) also says it cut Luna by 80% and Terra by 20%. The live post showed 20.9 million views when checked on 2026-09-13. It describes Fast mode for Sol as a speed option, not evidence that Standard pricing ended. Our 2026-09-11 interpretation of the pricing page was wrong. The conclusion outran the evidence. The prior article even cited the 2026-11-21 language, which should have stopped publication. Receipts only help when we listen to them. ## What the correction changes per 1,000 tasks For an 8k-input/2k-output workload, the current Standard bill is $72 per 1,000 Sol tasks, $40 for Terra, and $4 for Luna. The calculations are `(8,000 × input rate + 2,000 × output rate) / 1,000,000 × 1,000`. That makes Sol 18 times Luna's token bill for this shape: `$72 / $4 = 18`. This ratio says nothing about answer quality or retries. It only tells us what identical token volumes cost. Use the [LLM cost calculator](https://vynaris.com/calculator) when your workload differs. The [per-request cost](https://vynaris.com/glossary/category/cost-and-finops#per-request-cost) changes with the input-to-output ratio, cache share, and service tier. A rate card without a workload is accounting theatre. ## Cache and long-context math also move back down For Sol, 8,000 [cached input tokens](https://vynaris.com/glossary/category/inference-and-serving#prompt-caching) plus 2,000 output tokens cost `$3.20 + $40 = $43.20` per 1,000 Standard tasks. The previous article incorrectly gave $64. Current Sol long-context rates are $8 input and $30 output per MTok. A 300k-input/2k-output request therefore costs `(300,000 × $8 + 2,000 × $30) / 1,000,000 = $2.46`, not $3.09. Our [long-context pricing analysis](https://vynaris.com/blog/gpt-56-long-context-cliff-vs-gemini-200k-cost) explains when the higher rate applies. Batch remains half of Standard. On the same 8k/2k workload, [batch inference](https://vynaris.com/glossary/category/inference-and-serving#batch-inference) costs $36 per 1,000 Sol tasks: `(8,000 × $2 + 2,000 × $10) / 1,000,000 × 1,000 = $36`. Batch is cheaper because the request contract differs; it is not a free drop-in for synchronous traffic. ## What should change in a production budget Restore the current $4/$20 Sol, $2/$12 Terra, and $0.20/$1.20 Luna rows. At 100,000 monthly 8k/2k tasks, that means $7,200 for Sol, $4,000 for Terra, or $400 for Luna. Each total is the per-1,000 figure multiplied by 100. Do not treat 2026-11-21 as a guaranteed expiry date. OpenAI says the Sol price lasts *at least* through that date. It may continue longer. Keep a separate scenario for any future rate increase and re-fetch the provider page before changing routes. The [original Sol price-cut analysis](https://vynaris.com/blog/gpt-56-sol-price-cut-72-per-1000-balanced-tasks) had the right current rates and workload result. This update restores that basis. ## Honest tradeoff The corrected price strengthens the case for these models, but price alone cannot pick the model. A cheaper call can lose after retries, slow responses, or failed tasks. Measure cost per successful outcome before rerouting. There is also a documentation risk. Provider pricing pages combine service tiers, context bands, caching, and regional uplifts. Automating a diff without preserving section labels can manufacture a price change. We will retain the page snapshot with the row context from this correction forward. ## Sources - [OpenAI API pricing](https://developers.openai.com/api/docs/pricing), current rates, model IDs, service tiers, and Sol promotion note; verified 2026-09-13. - [OpenAI pricing announcement](https://x.com/OpenAI/status/2082878156483219672), published 2026-07-30; read directly in Firefox on 2026-09-13. - Reproducible arithmetic: `artifacts/gpt-56-promo-expired-sol-5-30-terra-2-5-15-luna-1-6-math.py`. ## About the author [Dhrumil Barot](https://vynaris.com/about#founder), Founder. Dhrumil is an AI inference and LLM infrastructure expert. Dhrumil built Vynaris to make model access, routing and pricing simple and transparent for developers. [LinkedIn](https://www.linkedin.com/in/bdhrumil) · [X](https://x.com/fyidhrumil)