In this blog post, we will see what OpenAI’s new GPT-5.6 family actually means for developers shipping AI features, not just what the press release says. OpenAI moved GPT-5.6 to general availability on July 9, 2026, and instead of one flagship model, it ships as three tiers: Sol, Terra, and Luna. I wanted to run a real workload through all three and see where each one actually earns its price tag.
GPT-5.6 launched on July 9, 2026, as three distinct capability tiers — Sol, Terra, and Luna — each with separate pricing and performance characteristics. The tiered structure allows OpenAI to update each tier independently and gives developers a cost-based routing option absent in previous model generations.
A customer support ticket triage task was used to benchmark all three tiers on throughput, time to first token, and cost. Luna delivered the fastest overall completion and lowest cost, Terra had the quickest time to first token, and Sol ranked last on both speed metrics despite carrying the highest price, suggesting flagship positioning applies to reasoning complexity rather than raw speed.
Table of Contents
- What Changed with GPT-5.6
- The One Example We Will Use Throughout
- Running the Benchmark with iamspeed.dev
- Reading the Results: Which Tier Fits Which Job
- Documenting the Run with achu.app
- Practical Takeaways for Teams
What Changed with GPT-5.6
The old naming scheme tied a model’s generation and its capability together. GPT-5.6 breaks that apart. The number marks the generation. Sol, Terra, and Luna mark durable capability tiers that can move forward on their own schedule going forward. That is a small naming decision with a real consequence: OpenAI can ship a Sol-only update without touching how Terra or Luna are served.
Here is the tier breakdown as of launch:
| Tier | Positioning | Input (per 1M tokens) | Output (per 1M tokens) |
|---|---|---|---|
| Sol | Flagship, long-horizon coding and research | $5.00 | $30.00 |
| Terra | Balanced, GPT-5.5 class quality at roughly half the cost | $2.50 | $15.00 |
| Luna | Fastest, cheapest, built for high-volume routine work | $1.00 | $6.00 |
Terra is worth pausing on. OpenAI is positioning it as the migration target for existing production workloads currently running on GPT-5.5, at about half the cost. Luna, meanwhile, still lands close to GPT-5.5 quality on several benchmarks despite being the cheapest tier in the family. That is the detail most people skim past and it is the one that actually changes architecture decisions.
The One Example We Will Use Throughout
Benchmarks are only useful when they map to something you would actually ship. I picked a workload most of us have already built some version of: customer support ticket triage. Given a raw support ticket, the model needs to do two things in one pass:
- Classify the ticket intent (billing, technical issue, account access, general question)
- Draft a short, ready-to-review reply
Here is the exact prompt I ran unchanged across Sol, Terra, and Luna:
You are a support triage assistant. Given the ticket below, respond with:
1. Intent category (billing, technical, account, general)
2. A two-sentence draft reply the agent can edit before sending
Ticket: "I was charged twice for my subscription this month and I can't find
an invoice for the second charge anywhere in my account."
Same prompt, same temperature, same system instructions, three models. That is the whole point. No tier gets an easier version of the task.
Running the Benchmark with iamspeed.dev
This is where I reach for my own tool. iamspeed.dev is the LLM streaming benchmarker I built specifically for this kind of comparison. It measures Time to First Token (TTFT) and tokens per second across providers, which matters more than raw benchmark scores the moment your feature is user-facing. A model that scores higher on a leaderboard but takes three extra seconds to start streaming will feel slower to a real user, and that gap does not show up in most published benchmarks.
iamspeed already supports the full GPT-5.6 family, so I lined up Sol, Terra, and Luna in a race and sent the same prompt to all three at once. Here is the actual result, as shown below:
Chick, running Luna, took the checkered flag with 155.6 tok/s. Sally, on Terra, actually had the fastest start at 1931ms TTFT. McQueen, on Sol, came in last on both counts, at 42.1 tok/s and a 4383ms TTFT.
Here is the full breakdown from that run:
| Tier | Model | Throughput | TTFT | Output Tokens | Total Time (TTLT) | Cost |
|---|---|---|---|---|---|---|
| Sol (McQueen) | gpt-5.6-sol | 42.1 tok/s | 4383ms | 210 tok | 9372ms | Highest |
| Terra (Sally) | gpt-5.6-terra | 79.2 tok/s | 1931ms (fastest start) | 383 tok | 6766ms | Mid |
| Luna (Chick) | gpt-5.6-luna | 155.6 tok/s (fastest overall) | 2570ms | 284 tok | 4394ms (fastest finish) | $0.00172 (cheapest) |
The detail that surprised me is that fastest start and fastest finish were two different models. Terra actually beats Luna out of the gate, its TTFT is nearly 25% quicker. But Luna’s throughput is almost double Terra’s, so it closes the gap and finishes the entire response first, while also landing as the cheapest request of the three. Sol never leads on speed or cost in this run, as shown below in the full race detail:
That is a useful thing to know before you assume “flagship” means “fastest.” In this run it did not.
Reading the Results: Which Tier Fits Which Job
If your feature streams a response to a user and total time to completion is what they actually feel, Luna is the one to reach for first. It won on throughput, won on total time, and won on cost in this run. For something like the ticket triage prompt earlier in this post, classification plus a short draft reply, that is squarely Luna’s territory.
Terra is worth keeping around for the moment TTFT specifically matters, cases where the user is staring at a blank screen and that first token showing up fast matters more than the total response time. A live chat widget where perceived responsiveness is everything is a good example.
Sol did not win a single metric in this particular race. That does not mean skip it entirely, it means Sol’s value shows up on harder reasoning tasks, not on a straightforward creative or classification prompt like the ones tested here. Running every request through Sol by default, based on this data, is paying flagship pricing for third place.
Documenting the Run with achu.app
Once you have a benchmark you actually want to keep, screenshots are usually the fastest way to share it, and this is where achu.app comes in. It is a screenshot beautifier I built with a Kovai Design Language look, and two features matter specifically for this kind of work:
- AI Bug Agent: useful if something in your triage pipeline breaks mid-benchmark and you want a quick annotated capture of what went wrong
- Privacy Guard OCR redaction: this one matters more than people expect. Benchmark screenshots often have API keys, request IDs, or actual customer ticket text sitting in plain view. Privacy Guard runs OCR over the screenshot and redacts that before you paste it into a report or a blog post
Head to achu.app if you are documenting a benchmark like this for a team wiki or a public writeup. It saves the manual blur-and-crop step most of us skip until after something leaks.
Practical Takeaways for Teams
If you are deciding where to start with GPT-5.6:
- Default new production workloads to Terra unless you have a specific reason not to
- Route high-volume, low-ambiguity tasks (classification, routing, short drafts) to Luna
- Reserve Sol for the tail of genuinely hard cases, not the whole pipeline
- Benchmark with your own prompts on iamspeed.dev before committing, since TTFT and throughput are workload-dependent
- Redact before you share, using something like achu.app‘s Privacy Guard, especially if your test prompts include real customer data
The tier split is a genuinely useful change, not just a rebrand. It gives you a cost lever that used to not exist: the same family, three honest price points, and a real reason to pick the one that matches the job instead of defaulting to the flagship out of habit.
Which tier are you defaulting to for production work right now? Let me know in the comments.
Happy Testing!