OpenAI GPT API pricing 2026 — GPT-5.5, 5.6 Sol, Terra & Luna costs, examples, cheaper access

This is OpenAI GPT API pricing for 2026: the per-model token rates for the GPT-5 family, worked cost examples, and the levers that cut a GPT bill the most. Kunavo prices the family about 65% below OpenAI's list rate behind one OpenAI-compatible API.

Rates last verified September 13, 2026. Kunavo's per-token prices below are read live from the model catalog, and the “OpenAI list” column tracks OpenAI's published rate — the official source is openai.com/api/pricing (developer detail at platform.openai.com/docs/pricing).

GPT API pricing at a glance

Rates are per 1M tokens, in USD, as billed on Kunavo. The “OpenAI list” column is OpenAI's published rate for the same model.

Live rates always show on the pricing page and each model page. GPT-5.6 Luna is the cheap workhorse, GPT-5.6 Terra the balanced default, and GPT-5.6 Sol, GPT-5.5 and GPT-6 Astra the heavy reasoners.

GPT-5.6 Sol is OpenAI's newest flagship — top-tier reasoning and agentic coding with a 1.05M-token context window. OpenAI lists it at $5.00 / $30.00 per 1M, the same list price as GPT-5.5, so on Kunavo it lands at the same $2.00 / $12.00 — about 60% under list. If you are already on gpt-5-5, moving to gpt-5-6-sol is a one-word change that costs the same per token.

One caveat on very long requests. OpenAI bills requests above 272,000 input tokens at 2× input and 1.5× output for the entire request, cached input included. Kunavo applies the same tier on GPT-5.5, GPT-5.6 Sol, Terra and Luna, and GPT-6 Astra, so past that threshold those models' rates double on input and rise by half on output — and the saving versus calling OpenAI directly is the same percentage on either side of it, because both prices move together.

How GPT token pricing works

You pay for input tokens (system prompt, context, messages) and output tokens (the completion). On the GPT-5 family output runs ~6× the input rate, so the biggest lever on cost is how much the model writes. These are reasoning models: the hidden reasoning tokens they generate are billed at the output rate, which is why capping max_tokens and lowering the reasoning effort matter.

Worked cost examples

Real numbers at Kunavo's rates:

So a million cheap Mini turns runs about $630. The math, runnable:

# Kunavo GPT rates (USD per 1M tokens): (input, output)
RATES = {
    "gpt-5-6-luna":  (0.070, 0.42),
    "gpt-5-6-terra":       (0.70, 4.20),
    "gpt-5-5":       (2.00, 12.00),
    "gpt-5-6-sol":   (2.00, 12.00),
}

def cost(model: str, in_tokens: int, out_tokens: int) -> float:
    i, o = RATES[model]
    # NOTE: on GPT-5 reasoning models, reasoning tokens are billed as output.
    return in_tokens / 1_000_000 * i + out_tokens / 1_000_000 * o

print(cost("gpt-5-6-luna", 1_000, 300))   # cheap turn   -> $0.00063
print(cost("gpt-5-6-terra", 6_000, 500))        # RAG answer   -> $0.009
print(cost("gpt-5-5", 20_000, 3_000))     # hard problem -> $0.076
print(cost("gpt-5-6-sol", 20_000, 3_000)) # newest flagship -> $0.076

Kunavo pricing and Stripe billing

No subscription, no OpenAI Platform billing setup. Top up a balance (Stripe or local payment methods) and calls draw down at the rates above. Pay-as-you-go from a $10 minimum top-up, the balance never expires, and larger top-ups carry bonus credit. The same balance covers GPT, Claude, Gemini, image, video and audio models — one wallet, one key.

Which GPT model should I choose?

• gpt-5-6-luna — default for high-volume chat, extraction and classification. Cheapest capable GPT tier.
• gpt-5-6-terra — the balanced default for most reasoning and RAG.
• gpt-5-6-sol — OpenAI's newest flagship and the default choice at the top of the range: strongest reasoning and agentic coding, 1.05M context, at the same per-token price as GPT-5.5.
• gpt-5-5 / gpt-6-astra — reach for these only when the cheaper tiers aren't accurate enough on the hardest problems. GPT-6 Astra is the deep-horizon tier, so justify it before you reach for it.

Route by difficulty — send the easy 80% to Mini, escalate the hard 20% — and see the AI cost optimization guide for the pattern in code. Comparing providers? See the Claude and Gemini pricing guides. Hitting ceilings rather than costs? OpenAI API rate limits covers which of RPM/TPM/RPD you actually hit and the retry that fixes it.

FAQ