OpenAI released GPT-6.1 Terra on October 26, filling the slot between GPT-6.1 Sol and GPT-6 Luna that GPT-5.6 Terra had held since July. That makes four tiers in the current lineup, with Terra below the GPT-6 Astra flagship and the Sol tier, aimed at visual reasoning, computer use, and high-volume work.
Independent benchmarks moved at the same time. Artificial Analysis's updated index widens the gap between Astra and Sol at the top and drops Luna well below the other three, which changes where each tier earns its cost.
The tier names still hide how much the lineup overlaps. Sol at max matches Astra one effort level down, and Terra at its lowest setting outscores Luna at its highest, so the real choice is a tier and an effort level for every step of an agent.
So the question with GPT-6.1 Sol vs Terra vs Luna vs Astra is which one to use for each step in 2026.
What are GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra, and GPT-6 Luna?
GPT-6 Astra is the flagship of the current GPT-6 family, GPT-6.1 Sol sits one tier below it, GPT-6.1 Terra below Sol, and GPT-6 Luna is the entry tier. All four share a 1.05M-token context window and a 128K-token output limit, and OpenAI says Luna was trained with methods similar to Astra's.
GPT-6.1 Sol
GPT-6.1 Sol is OpenAI's upgrade to GPT-6 Sol, released a week after it, though OpenAI hasn't deprecated the older model and gpt-6-sol is still available in the API. It costs $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens, and its reasoning effort runs from low to max, with medium as the default and no none setting.
The names are where it gets confusing. In GPT-5.6, Sol was the frontier tier, with Terra in the middle and Luna at the bottom. GPT-6 puts Astra above Sol, so the same names now sit one rung lower, and gpt-6.1-sol is a separate model from both gpt-6-sol and gpt-5.6-sol, all three still available alongside the rest of the GPT models OpenAI hasn't retired.
GPT-6.1 Terra
GPT-6.1 Terra is the newest GPT-6 tier, released October 26 as the replacement for GPT-5.6 Terra. OpenAI pitches it for visual reasoning, computer use, and natural writing, with examples like matching product photos to catalog entries or acting on application screens.
It costs $1 per million input tokens, $0.05 cached, and $5 output, half of 6.1 Sol's rates. Its reasoning effort runs from none to max with medium as the default, and its knowledge cutoff is June 1, 2026.
GPT-6 Luna
GPT-6 Luna is the GPT-6 model OpenAI builds for focused, high-volume tasks, where each call is short and the output is easy to check.
At $0.10 input and $0.50 output per million tokens, Luna costs one tenth of Terra, one twentieth of 6.1 Sol, and one hundredth of Astra on both lines. Its reasoning effort runs from none to max, a range it shares only with Terra.
GPT-6 Astra
GPT-6 Astra sits above the other three as the model OpenAI positions for the hardest end-to-end work. At $10/$50 per million tokens it costs five times 6.1 Sol's price, and its effort range runs from low to max, the same as 6.1 Sol.
GPT-6.1 Sol vs Terra vs Luna vs Astra: overview
All four GPT-6 models accept text and images and return text, with the same context window and output limit. The differences are in price and effort range.
| GPT-6 Astra | GPT-6.1 Sol | GPT-6.1 Terra | GPT-6 Luna | |
|---|---|---|---|---|
| API model ID | gpt-6-astra | gpt-6.1-sol | gpt-6.1-terra | gpt-6-luna |
| Input / output per 1M | $10 / $50 | $2 / $10 | $1 / $5 | $0.10 / $0.50 |
| Cached input per 1M | $1.00 | $0.10 | $0.05 | $0.01 |
| Context window | 1.05M tokens | 1.05M tokens | 1.05M tokens | 1.05M tokens |
| Max output | 128K tokens | 128K tokens | 128K tokens | 128K tokens |
| Reasoning effort | low to max | low to max | none to max | none to max |
| Knowledge cutoff | Apr 30, 2026 | Apr 30, 2026 | Jun 1, 2026 | May 18, 2026 |
| Best for | Fast answers on hard steps | Default for most agent steps | Visual and computer-use steps at volume | Cheapest checkable calls |
Because the spec sheets barely differ, choosing between the four comes down to what each tier delivers per dollar at a given effort level.
GPT-6.1 Sol vs Terra vs Luna vs Astra: pricing
Output tokens cost five times the input rate on all four GPT-6 tiers, and reasoning tokens bill as output. As a result, the effort level you pick often moves the bill more than the tier's headline input price, and effort defaults differ across the best reasoning models from other labs, from medium on Claude Opus 5.5 to max on Kimi K3.
Per-token pricing
On standard input and output, Astra costs five times 6.1 Sol, Sol costs twice Terra, and Terra costs ten times Luna. Compared with GPT-5.6, 6.1 Sol and Luna come in at exactly half of the previous promotional rates of $4/$20 and $0.20/$1.20, and OpenAI puts Terra's standard rates at half of GPT-5.6 Terra's.
The service-tier multipliers are the same across the family. Batch and Flex are half of Standard, Fast mode is double, and regional processing adds 10% where it's available. EU data residency narrows those options differently by model: 6.1 Sol can't use Fast mode with EU residency, and Luna only supports Standard processing with it, so an EU Luna workload can't pair residency with the Batch or Flex discount.
Caching and long-context pricing
Cached input bills at 10% of the uncached rate on Astra and Luna and at 5% on 6.1 Sol and Terra, and cache writes bill at 1.25x on all four. Caching also shapes routing by step, because a cached prefix belongs to the model that processed it, so moving a conversation from 6.1 Sol to Astra starts the cache over.
Long prompts have their own threshold. Once input passes 272K tokens, the whole request bills at 2x the input and cache rates and 1.5x the output rate, so a 300K-token 6.1 Sol request pays $1.20 for input alone where a 270K-token request pays $0.54. Though, that threshold differs across the other LLM APIs. If most of that context repeats between calls, prompt caching is the first place to recover it.
Cost per task
Per-token prices don't account for how many tokens each model spends. On the Artificial Analysis Intelligence Index at max effort, 6.1 Sol uses about 68K output tokens per task against Astra's 52K.
Astra at max costs $6.24 per task against 6.1 Sol's $1.27, close to the five-times gap on the rate card, while Terra at max costs $0.96 and Luna $0.15. Terra's gap to Sol is narrower than its half-price rates suggest at max, at three quarters of Sol's cost per task, but it widens at lower settings, where Terra at medium costs $0.17 against Sol's $0.36.
Effort level stretches the range further. 6.1 Sol costs $0.16 per task at low and $1.27 at max, about eight times as much inside one model.
GPT-6.1 Sol vs Terra vs Luna vs Astra: performance
Astra leads at every effort level, by two to eight points over 6.1 Sol on Artificial Analysis's current index. Terra lands between Sol and Luna, and Luna sits well below the other three.
Intelligence and coding
On version 5.0 of the Intelligence Index, a composite of nine evaluations, Astra scores 57 at max effort, 6.1 Sol 55, Terra 51, and Luna 28. The previous version scored Astra 53, Sol 52, and Luna 37, but version 5.0 adds new coding and science evaluations, so scores don't compare across the two.
Coding is where the middle tiers close the gap. On Terminal-Bench 4.0, Astra scores 59% at max while 6.1 Sol and Terra both score 56%, and on FrontierCode 2.0, which grades whether a maintainer would merge the agent's change, Astra and Terra score 42% and Sol 41%.
Knowledge work
Knowledge work is where Terra's lower overall score comes from. On AA-Briefcase, a long-horizon knowledge work evaluation, Astra and 6.1 Sol score 1569 and 1555 Elo against Terra's 1402, and on HLE-Diamond, a reasoning and knowledge test, Terra scores 47% against Sol's 59% and Astra's 64%. Terra's AA-Omniscience net score of 30 also trails the 42 that Astra and Sol both post.
For a step whose output someone reads and acts on, such as a migration plan or a written report, that gap is a reason to score Terra, and Luna even more so, against your own rubric before moving the step down a tier. Visual work is the other direction, since OpenAI reports Terra nearly matching Astra and 6.1 Sol on MMMU-Pro, a visual reasoning benchmark, at a much lower cost.
Speed and latency
Output speed rises as the tiers get cheaper. Across effort levels, Astra generates 54 to 72 output tokens per second, 6.1 Sol 61 to 87, Terra 91 to 109, and Luna 116 to 128.
Effort level drives time to first token far more than tier does. At low effort, every tier starts answering in about 1.5 to 3 seconds, and at medium Astra takes about 6 seconds, Sol 9, and Terra 5. From high effort up the waits stretch: Astra takes 107 seconds at high and 368 at max, Sol 70 and 331, and Terra 21 and 261, while Luna reaches 122 at max.
For any step a user waits on, low or medium effort on any tier keeps the delay in single-digit seconds, and Terra at high is the one setting above medium that stays near 20.
Performance by effort level
The same benchmarks score each model at every effort level from low to max, including GPT-6 Luna and GPT-6.1 Terra, though Artificial Analysis doesn't publish scores for the none setting on Terra or Luna.
| Effort | GPT-6 Astra | GPT-6.1 Sol | GPT-6.1 Terra | GPT-6 Luna |
|---|---|---|---|---|
| max | 57 · $6.24 | 55 · $1.27 | 51 · $0.96 | 28 · $0.15 |
| xhigh | 55 · $3.66 | 52 · $0.76 | 45 · $0.40 | 23 · $0.10 |
| high | 55 · $3.24 | 50 · $0.59 | 42 · $0.26 | 21 · $0.07 |
| medium | 51 · $2.73 | 46 · $0.36 | 38 · $0.17 | 18 · $0.04 |
| low | 43 · $1.42 | 35 · $0.16 | 29 · $0.07 | 10 · $0.01 |
Intelligence Index v5.0 score · cost per task
Sol at max ties Astra at high (55) for $1.27 against $3.24, and Sol at xhigh (52) outscores Astra at medium (51) for $0.76 against $2.73. Astra's own xhigh setting adds nothing over high on this index, at 55 for both, while costing more and taking twice as long to start answering.
Terra fills in the lower half of that ladder. Terra at max matches Astra at medium (51) for about a third of the cost per task, and Terra at low (29) already outscores Luna at max (28) for about half the cost, starting its answer in 1.5 seconds instead of two minutes.
What you give up is latency at the top. Sol at max takes about five and a half minutes to its first token and Terra at max over four, where Astra at medium reaches 51 and starts answering in about 6 seconds.
Tool calling adds one more constraint in Chat Completions. 6.1 Sol and Terra need the Responses API for any tool call, and Luna supports function calling in Chat Completions only at none.
Which is best: Sol, Terra, Luna, or Astra?
No single GPT-6 model is the right choice for a whole agent. Astra leads at every effort level, but 6.1 Sol reaches Astra's scores one level down for under half the cost, and Terra does the same against Luna, so the useful question is which tier and effort to use for each kind of step.
Pick a tier and an effort level
Across the benchmark data, a handful of configurations cover most agent steps:
- GPT-6.1 Sol at medium - The default for general agent work, tool calling, and code edits. It scores 46 for $0.36 per task and returns its first token in about 9 seconds.
- GPT-6.1 Sol at high or xhigh - For hard steps nobody is waiting on, like planning a multi-file change in a background job. It scores 50 to 52 for $0.59 to $0.76 per task, up to a point above Astra at medium for under a third of the cost.
- GPT-6.1 Sol at max - For steps that need Astra-level scores and can wait several minutes. It ties Astra at high (55) for about 40% of the cost per task.
- GPT-6 Astra at medium - For a hard step that also needs a fast first token, like recovering after a tool call fails mid-conversation. It scores 51 and starts answering in about 6 seconds.
- GPT-6 Astra at max - For the few steps where two more points matter more than cost. It scores 57 for $6.24 per task.
- GPT-6.1 Terra at low or medium - For extraction, classification, screenshot reading, and other high-volume steps with an answer you can check. Terra at low scores 29 for $0.07 per task with a 1.5-second first token, and Terra at medium scores 38 for $0.17.
- GPT-6 Luna at low - For the simplest checkable calls, like routing or yes-or-no classification, where Luna scores 10 for $0.01 per task. That cost floor is the tradeoff at the center of any Haiku vs Luna decision.
The setting to question is Astra at xhigh, which scores the same 55 as Astra at high for more cost per task and roughly twice the wait to its first token.
Route each agent step to a different tier
Every call in an agent run can set its own model and effort, so the tier decision can happen per step instead of once per application. A coding agent might triage an incoming issue with Terra at low, plan the change with 6.1 Sol at xhigh, make the edits with Sol at medium, check a screenshot of the failing UI with Terra at medium, and write the pull request summary with Terra again. If a user is waiting on the plan, Astra at medium gets a comparable score back in seconds.
Where those boundaries go depends partly on caching. Switching models restarts the prompt cache, so the split works best at natural handoffs, such as passing a finished plan to the step that executes it. Tool calling adds one more constraint, since every tool call on 6.1 Sol or Terra, and any tool call on Luna above none, needs the Responses API.
A fixed split like this is simpler to reason about than an LLM router that picks a model per request from its content. Each step's assignment is also a single setting you can score on your own traffic, with a rubric-based evaluator such as LLM-as-a-judge.
Route GPT-6.1 Sol, Terra, Luna, and Astra with Respan
Routing agent steps across GPT-6 tiers pays off when switching models takes one change and every call's cost is visible afterward. Respan lets you route, observe, and evaluate every LLM call, with discounted rates on several GPT-6 routes listed on each model's page.
Respan's LLM gateway puts GPT-6.1 Sol, GPT-6.1 Terra, GPT-6 Luna, and Astra behind one OpenAI-compatible endpoint with 1,000+ models, so moving a step to a different tier means changing one word in the request.
- Fallback when a provider fails - Define a fallback chain once, at the org level or per request, and traffic moves to the next model in the chain when a provider errors or rate-limits.
- Cost for every step - Each request is logged as a span with latency and cost attached, and spend breaks down by model, request, and end customer, with hard limits that stop spend at a ceiling.
- Traces across the whole run - Every LLM call, tool run, and agent turn nests in one trace, so a bad output leads back to the step and model that produced it.
- Evals before and after a switch - Build datasets from production logs, run them through each model with your evaluators, then deploy the same evaluators on live traffic so a regression after moving a step to a cheaper tier surfaces in real time.
FAQ
Is GPT-6.1 Sol the same as GPT-6 Sol?
No. gpt-6.1-sol is a separate model from gpt-6-sol, with the same $2/$10 per million tokens but cached input at $0.10 instead of $0.20. It drops the none effort setting, running from low to max, and Artificial Analysis scored it four points above GPT-6 Sol at max effort when it launched.
Is there a GPT-6 Terra?
Yes, under the name GPT-6.1 Terra. OpenAI released gpt-6.1-terra on October 26, 2026 as the replacement for GPT-5.6 Terra, and it sits between 6.1 Sol and Luna at $1 input and $5 output per million tokens. On Artificial Analysis's index it scores 51 at max effort, against 55 for 6.1 Sol and 28 for Luna.
Which ChatGPT plans include GPT-6.1 Sol, Terra, and Luna?
GPT-6.1 Sol and GPT-6 Luna are in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, and Free and Go users can use Luna in the desktop app. GPT-6.1 Terra is in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu users.
Does GPT-6.1 Sol support tool calling in Chat Completions?
No. OpenAI's GPT-6.1 Sol model page supports Chat Completions only without tool calling, so any tool call needs the Responses API, and GPT-6.1 Terra has the same requirement. GPT-6 Luna supports tool calls in Chat Completions only at none effort, and Astra's model page doesn't list the same restriction.
Can I use GPT-6.1 Sol, Terra, Luna, and Astra in the same app?
Yes. Through the Respan gateway, all four sit behind one endpoint, so each call can name its own GPT-6 model and effort level, and every call's cost and trace land in one place. OpenAI's API also accepts any of the four on each request, since they share the same endpoints.
