GPT-5.6 Sol Is Now 50% Cheaper — Here's What That Actually Means for Developers

TL;DR

GPT-5.6 Sol is currently listed by OpenAI at $5 per million input tokens and $30 per million output tokens. OpenRouter also currently lists Sol at $5/$30.

On July 30, 2026, OpenAI announced a 20% price reduction for Terra and an 80% reduction for Luna, with the announcement giving new prices of $2/$12 for Terra and $0.20/$1.20 for Luna. However, OpenAI's current API model documentation now lists Terra at $2.50/$15 and Luna at $1/$6.

That discrepancy matters. For a production cost model published on August 18, the live model pages are the safer source for current pricing, while the July 30 figures should be treated as the prices announced on that date rather than silently substituted for today's rate card.

The main takeaway is simple: Sol did not get a 50% price cut. The important developer decision is how much work should go to Sol versus the cheaper Terra and Luna tiers.

Sources: GPT-5.6 Sol, GPT-5.6 Terra, GPT-5.6 Luna, OpenAI July 30 update, OpenRouter GPT-5.6 Sol

What Actually Happened — And What Didn't

The first thing to correct is the headline that started this discussion: GPT-5.6 Sol is not currently listed at $2.50/$15.

OpenAI's current Sol model page lists:

• Input: $5.00 per 1M tokens

• Cached input: $0.50 per 1M tokens

• Output: $30.00 per 1M tokens

OpenRouter's current Sol listing also shows $5/$30 per 1M tokens.

So the claim that Sol itself became 50% cheaper does not match the live model pages.

The more interesting part of the story is what happened on July 30. OpenAI published a pricing update announcing lower prices for Terra and Luna while leaving Sol's price unchanged.

According to that July 30 announcement:

• GPT-5.6 Terra: $2 input / $12 output

• GPT-5.6 Luna: $0.20 input / $1.20 output

• GPT-5.6 Sol: unchanged

But OpenAI's current model pages now show:

• GPT-5.6 Terra: $2.50 input / $15 output

• GPT-5.6 Luna: $1 input / $6 output

• GPT-5.6 Sol: $5 input / $30 output

That means an article published today should not present the July 30 figures as the current live rate card without qualification.

The Current GPT-5.6 Family Pricing

For the current API model pages, the pricing picture is:

All prices are per million tokens.

At today's listed rates, Sol costs 2× Terra on both input and output and 5× Luna on both.

That is still a meaningful gap, but it is different from the 25× spread implied by the July 30 announcement prices.

OpenAI also documents a long-context multiplier for GPT-5.6: prompts containing more than 272,000 input tokens are charged at 2× the normal input price and 1.5× the normal output price for the full request.

For caching, OpenAI says cache writes are billed at 1.25× the model's uncached input rate, while cached reads receive a 90% discount.

Sources: Sol, Terra, Luna

What GPT-5.6 Sol Actually Is

GPT-5.6 Sol is the frontier model in the GPT-5.6 family, aimed at complex professional work, reasoning, coding, and demanding agentic workflows.

OpenAI lists:

• 1.05 million-token context window

• 128,000 maximum output tokens

• Reasoning controls from none through max

• Responses API and Chat Completions support

• Tool support including web search, file search, computer use, and code interpreter

The gpt-5.6 alias routes to GPT-5.6 Sol.

OpenAI's own model guidance is straightforward:

• Sol: frontier capability for complex reasoning and coding

• Terra: balance of intelligence and cost

• Luna: efficient, high-volume workloads

Sources: GPT-5.6 Sol, OpenAI model guidance

The Real Cost Question: Which Requests Need Sol?

The useful question is not "How can I get Sol at half price?"

It is:

Which requests actually need the flagship model?

At the current live prices, a simple three-tier routing strategy makes sense:

Luna is the candidate for high-volume, cost-sensitive work such as classification, extraction, simple transformations, and routine automation.

Terra is the middle tier for everyday coding, analysis, and agentic workloads where you want more capability without paying the flagship rate.

Sol is the escalation tier for difficult reasoning, complex coding, long-horizon tasks, and workloads where a higher success rate or better tool use justifies the additional spend.

That is an architectural decision rather than a simple model-price comparison. A router that sends routine requests to Luna or Terra can keep Sol available for the requests where it produces measurable value.

What About Batch Processing?

GPT-5.6 Sol supports the Batch API, but OpenAI's current Sol model page still lists the model itself at $5/$30.

That is why you should avoid publishing a blanket statement such as "Sol is $2.50/$15 through Batch" unless the current official pricing documentation explicitly shows that exact price for your workload.

Batch remains valuable for asynchronous work because it is designed for workloads that do not require immediate responses. Good candidates include:

• Bulk document analysis

• Code-review queues

• Offline classification

• Evaluation runs

• Nightly reports

• Large content-generation pipelines

The safest cost-modeling rule is simple: check the live Batch pricing for the exact model and account before assuming a 50% discount.

Source: GPT-5.6 Sol API documentation

OpenRouter: Useful for Routing, Not a Guaranteed Sol Discount

OpenRouter can be useful when you want a single OpenAI-compatible interface across multiple providers.

Its current GPT-5.6 Sol page lists:

$5 input / $30 output per 1M tokens.

OpenRouter also notes that its effective pricing can be lower than provider list pricing when prompt caching is involved, and that it can route requests across providers based on routing modes.

That means OpenRouter's strongest argument here is routing and provider flexibility, not a guaranteed 50% Sol discount.

Do not hard-code a $2.50/$15 Sol assumption into a budget model simply because a marketplace or older article quoted that number. Check the live listing immediately before calculating your production economics.

import OpenAI from "openai";

const client = new OpenAI({
  baseURL: "https://openrouter.ai/api/v1",
  apiKey: process.env.OPENROUTER_API_KEY,
});

const response = await client.chat.completions.create({
  model: "openai/gpt-5.6-sol",
  messages: [
    { role: "user", content: prompt }
  ],
  max_tokens: 2048,
});

console.log(response.choices[0].message);

Source: GPT-5.6 Sol on OpenRouter

The Pricing Story Nobody Should Miss

The important story is not that Sol suddenly became a mid-tier model.

It didn't.

Sol remains the flagship and remains priced accordingly.

The bigger story is that OpenAI is maintaining a three-tier capability ladder:

graph LR
    SOL["Sol<br/>$5 / $30<br/>Frontier"]
    TERRA["Terra<br/>$2.50 / $15<br/>Balanced"]
    LUNA["Luna<br/>$1 / $6<br/>Cost-sensitive"]

    SOL -->|"2×"| TERRA
    TERRA -->|"2.5×"| LUNA
    SOL -->|"5×"| LUNA

The exact July 30 announcement was more aggressive:

graph LR
    SOL["Sol<br/>$5 / $30<br/>Price unchanged"]
    TERRA["Terra<br/>$2 / $12<br/>Announced Jul 30"]
    LUNA["Luna<br/>$0.20 / $1.20<br/>Announced Jul 30"]

    SOL -->|"2.5×"| TERRA
    TERRA -->|"10×"| LUNA
    SOL -->|"25×"| LUNA

That second diagram is useful as historical context, but the first diagram is the one developers should use for the current live model pages.

What This Changes for Your Stack

If you're running an older GPT model for routine work: benchmark Terra first. OpenAI explicitly positions it as the balance between intelligence and cost.

If you're running very high-volume workloads: benchmark Luna. The current API page lists it at $1/$6, making it the family's lower-cost option.

If you're building complex coding or reasoning systems: keep Sol as the premium escalation tier. Its value comes from capability, not from a discounted token price.

If you're building agentic systems: treat reasoning effort and model selection as separate cost levers. GPT-5.6 supports multiple reasoning levels, so the cheapest architecture is not necessarily "always use the cheapest model"; it is "use the least expensive configuration that reliably completes the task."

The Practical Checklist

Before changing your production cost model:

☐ Use the current OpenAI model page as the source of truth for live model prices

☐ Treat GPT-5.6 Sol as $5/$30 unless the official pricing page says otherwise

☐ Do not mix the July 30 announced Terra/Luna prices with today's live model-page prices

☐ For current planning, use Terra at $2.50/$15 and Luna at $1/$6

☐ Check whether requests exceed 272K input tokens before estimating long-context costs

☐ Measure cached versus uncached input before forecasting token spend

☐ For Batch, verify the current model-specific pricing instead of assuming a universal 50% discount

☐ If using OpenRouter, check the live provider/model listing before building a budget

☐ Route routine workloads downward and reserve Sol for tasks that genuinely benefit from its additional capability

Bottom Line

The original "Sol is now 50% cheaper" story is not supported by the current live model pages.

GPT-5.6 Sol is still $5/$30.

The more useful developer story is that GPT-5.6 has a clear capability ladder, OpenAI announced major Terra and Luna reductions on July 30, and the current model pages now show a different live price state for those two models.

For developers, that means the winning strategy is not chasing a headline discount. It is routing each workload to the cheapest GPT-5.6 tier that reliably gets the job done.

Further Reading

• GPT-5.6: Frontier intelligence that scales with your ambition — OpenAI

• Advancing the price-performance frontier with GPT-5.6 — OpenAI, July 30, 2026

• GPT-5.6 Sol API documentation

• GPT-5.6 Terra API documentation

• GPT-5.6 Luna API documentation

• GPT-5.6 model guidance

• GPT-5.6 Sol on OpenRouter

Pricing can change quickly. Before updating a production budget, confirm the live price on the official OpenAI model page and the exact provider listing you plan to use.