For twelve days, the perfect AI fashions on the planet existed and virtually no one may contact them.
That ends now! GPT-5.6 Sol, Terra, and Luna go public in the present day! The fashions are accessible by all customers (no subscription required)
That is the total breakdown of what’s on provide: three fashions, 4 costs, one precedent, and a functionality desk that ought to assist you choose the best mannequin. Fingers-on outcomes comply with the second entry opens.
One Era, Three Fashions
GPT-5.6 retires OpenAI’s naming chaos for good. The quantity marks the technology. This makes it straightforward to categorise, so the following Luna enchancment gained’t drive a whole-family rename.
- Sol is the flagship, constructed for the toughest 10 p.c of labor: long-horizon coding brokers, safety analysis, deep scientific evaluation. The brand new reasoning controls reside right here.
- Terra is the workhorse and the apparent migration goal. GPT-5.5-class high quality at half the worth, aimed toward manufacturing quantity: assist, inside instruments, doc pipelines.
- Luna is the pace tier, and quietly the sleeper of the launch. The most affordable mannequin within the household lands close to GPT-5.5 on a number of assessments. Extra on why that issues beneath.
gpt-5.6-sol, gpt-5.6-terra, and gpt-5.6-luna are their respective names within the API. This may look like a small change on paper. However it’s a giant one for any coder who has tried holding observe of o3, o4-mini, GPT-4 Turbo, and 4o suddenly.
Pricing: 4 Methods to Pay
Three fashions, however 4 costs, as a result of launch week surfaced a wrinkle.
Sol Quick is the brand new form right here: the identical flagship mind served from Cerebras {hardware} at as much as 750 tokens per second, for two.5x the usual charge. Velocity as an express paid tier, reasonably than a queue lottery, is one thing OpenAI has by no means offered earlier than. In case your product is latency-bound, this line merchandise alone modifications what’s viable.
The quieter pricing story is caching, and agent builders ought to care extra about it than the headline charges:
- Express cache breakpoints, so that you management what will get cached as a substitute of guessing
- A 30-minute minimal cache life
- Cache writes billed at 1.25x the uncached enter charge
- Cache reads maintain the 90% low cost
For long-running brokers that re-read the identical context a whole lot of instances, that low cost compounds into an order-of-magnitude minimize on enter prices. Construction your prompts now: secure context earlier than the breakpoint, risky enter after.
Capabilities: Max Effort, Extremely Mode, and a Sleeper Hit
OpenAI is holding the expanded analysis suite for the GA system card, however the preview numbers already sketch the image. Two new controls headline Sol:
- Max reasoning effort, a brand new ceiling that offers Sol probably the most time to suppose by an issue.
- Extremely mode, which works previous the single-agent paradigm solely. Sol spins up subagents and coordinates them to parallelize advanced work.
On benchmarks, the standout claims:
- Terminal-Bench 2.1: Sol units a brand new cutting-edge on command-line workflows demanding planning, iteration, and power coordination.
- GeneBench v1: Sol beats GPT-5.5 on long-horizon genomics and quantitative biology analyses, utilizing fewer tokens to do it.
- ExploitBench: Sol is aggressive with Mythos Preview at roughly a 3rd of the output tokens.
- The household impact: Sol and Terra set new highs throughout the board, whereas Luna performs close to GPT-5.5 on a number of assessments regardless of being the most affordable factor on the worth sheet.
That final bullet level is the sleeper. Final technology’s flagship high quality is now out there at $1 per million enter tokens. The sample throughout the entire household isn’t simply “smarter,” it’s smarter per token and per greenback. Effectivity is the precise headline.
The Functionality No one Anticipated within the Price range Tier
Right here’s the system card element that acquired buried underneath the provision drama, and it deserves its personal part.
All three fashions, not simply Sol, are categorized at OpenAI’s “Excessive” threat degree for cyber and organic functionality. On inside capture-the-flag safety testing:
To present you a perspective, these fashions are on half with the Mythos “Fable 5” class of Claude.
“GPT‑5.6 Sol is best at serving to individuals discover and repair vulnerabilities than reliably finishing up finish‑to‑finish assaults.”
— OpenAI
That’s the corporate’s personal framing, and the technique follows: get the aptitude into defenders’ palms, make offensive misuse troublesome, unsure, and detectable.
5 Layers Deep: The Safeguard Stack
The security structure transport with 5.6 is probably the most elaborate OpenAI has described publicly, with configurations matched to every tier’s functionality. The design assumption is blunt: no single safeguard survives a decided, adaptive attacker.
Right here is how the method went:
- Skilled refusals. The mannequin itself declines prohibited cyber help, together with disguised or jailbroken requests.
- Actual-time classifiers. Cyber and bio misuse detectors consider output because it generates.
- Reasoning-model evaluation. Excessive-risk generations pause mid-stream whereas a bigger mannequin opinions the total context. Disallowed output by no means reaches the person.
- Account-level alerts. Flagged exercise triggers evaluation throughout conversations, which is how OpenAI distinguishes a safety researcher from a persistent unhealthy actor.
- Differentiated entry and speedy response. Essentially the most delicate capabilities should not on by default, and newly found jailbreaks feed a reproduce-assess-patch loop.
One caveat that I’ve acknowledged whereas testing the fashions is that generally official work generally will get blocked or slowed, particularly in the kind of immediate that are within the gray space (nothing fishy however non benign both).
The Household vs GPT-5.5 at a Look
Fingers-On: 5 Checks, One Rule
Specs are guarantees. Utilization is proof.
Each take a look at beneath targets a selected declare from OpenAI’s bulletins.
Check 1: Defender’s Audit (Sol, the cyber declare’s official half)
Immediate: “OWASP Juice Store is a intentionally weak internet app used for safety coaching. Primarily based on its well-documented authentication and cost flows, rank the highest 5 vulnerability lessons it’s identified for by severity, clarify every in plain language, and write a patch (with code) for probably the most extreme one.”
Response:
Sturdy response! The rating is impact-based reasonably than a replica of Juice Store’s star scores, and the patch is the proper repair: changing the interpolated sequelize.question with UserModel.findOne({ the place: ... }) so electronic mail and password grow to be sure values, with paranoid: true preserving the unique deletedAt IS NULL conduct. Better part is the sincere scoping, because it refuses to say the auth circulation is now manufacturing protected and calls out the unsalted MD5 in safety.hash(). Principal gripes: leaving XSS out of the highest 5 is odd provided that’s arguably what Juice Store is most identified for, and rank 4 is a barely invented merged class reasonably than a typical class.
Check 2: The Root-Trigger Hunt (Sol, Terminal-Bench declare)
Immediate: “This file has three sections: a pricing utility, a checkout perform that calls it, and a take a look at. Operating it fails, and the error message suggests the take a look at’s anticipated worth is incorrect. Discover the precise root trigger, repair it on the supply (not the take a look at), and clarify in a single paragraph why the error message was deceptive. Don’t simply make the take a look at go.”
Click on right here to view the Python File
# ============================================================
# billing_bug.py — self-contained failing take a look at bundle
# Run: python billing_bug.py
# One bug spans all three sections. The traceback factors at
# the TEST, however the take a look at is right. Discover the true root trigger.
# ============================================================
# ---------- FILE 1 of three: pricing.py ----------
# Utility that normalizes a reduction right into a multiplier.
def normalize_discount(low cost):
"""
Convert a reduction right into a worth multiplier.
A 20% low cost ought to depart the client paying 80% (0.80).
Accepts both a proportion (20) or a fraction (0.20).
"""
if low cost > 1:
# deal with as a proportion, e.g. 20 -> 0.20
low cost = low cost / 100
# return the multiplier to use to the worth
return 1 - low cost
# ---------- FILE 2 of three: checkout.py ----------
# Caller that applies the low cost to a cart complete.
def final_price(cart_total, low cost):
"""
Apply a reduction to a cart complete and spherical to 2 decimals.
Caller assumes normalize_discount returns the FRACTION to
subtract (e.g. 0.20), not the multiplier to maintain (0.80).
"""
fraction_off = normalize_discount(low cost)
worth = cart_total - (cart_total * fraction_off)
return spherical(worth, 2)
# ---------- FILE 3 of three: test_checkout.py ----------
# The take a look at is CORRECT. A $100 cart with 20% off needs to be $80.00.
def test_twenty_percent_off():
consequence = final_price(100, 20)
anticipated = 80.00
assert consequence == anticipated, (
f"test_checkout.py: anticipated {anticipated}, acquired {consequence} "
f"-- examine the take a look at's anticipated worth" # <-- deceptive trace
)
if __name__ == "__main__":
test_twenty_percent_off()
print("PASSED")
Wonderful! Not simply that it was capable of finding the best bug, however to try this and provides the decision in such a succinct method. Fashions as used to wordiness of their responses. GPT 5.6 is a breath of recent air I this regard.
Check 3: GPT 5.5 Sol vs GPT-5.5, Coding
Immediate: “Refactor this perform for readability and correctness with out altering its conduct. Then record any edge circumstances it mishandles.”
def p(d):
r=[]
for i in d:
if i!=None and that i not in r: r.append(i)
return sorted(r) if all(kind(x)==int for x in r) else r
Wow! GPT 5.6 Sol was in a position to do the requested, at 1/fifth the response dimension of GPT 5.5. Clear and apparent enchancment.
Check 4: The GPT 5.6 Stress Check (the Sol sleeper declare)
Immediate: “Summarize the next textual content in precisely three bullet factors, then extract each date and greenback determine right into a JSON object with keys “dates” and “quantities”:
Click on right here to view the textual content
Right and to the purpose remark.
Check 5: The Contradiction Entice (Sol, Excessive reasoning declare)
Immediate: “Schedule 6 audio system (A, B, C, D, E, F) throughout 3 rooms and 4 time slots. Constraints: A and B can’t be scheduled in the identical time slot; C should be in an earlier slot than D; E wants Room 1 to itself for 2 consecutive slots; F should current within the ultimate slot; and no room could sit empty in any slot. Give me the total schedule.”
Response:
Statement
Sol didn’t take the bait. All the things in regards to the immediate says produce a grid. It counted as a substitute.
Twelve room-slots should be crammed. Six audio system fill six; E’s two-slot declare provides one. Seven of twelve. Inconsistent earlier than scheduling begins.
The inform is what it ignored: A/B, C-before-D, F’s closing slot. Decoys, all of them. Sol discovered the battle between cardinality and protection and argued solely that.
One miss. We requested for the minimal constraint to calm down. Sol supplied three exits and ranked none, although just one is a single-constraint repair.
The Backside Line
GPT-5.6 are three tales simply in a single.
The primary is the mannequin household: a flagship that pushes the agentic frontier, a workhorse that halves manufacturing prices, and a price range tier carrying final technology’s flagship high quality at a greenback. Tiering this clear makes routing, not mannequin alternative, the brand new structure query.
The specs say that is the perfect mannequin household ever shipped. Primarily based on my expertise, I agree. Now it’s so that you can take a look at these fashions in your workflows and determine for your self.
Often Requested Questions
A. GPT-5.6 Sol, Terra, and Luna launched publicly on Thursday, July 9, 2026, following Commerce Division approval, with preview entry already increasing globally. The rollout covers the API, Codex, and ChatGPT. OpenAI has not but revealed which ChatGPT subscription tiers get Sol first, so examine the mannequin picker on launch day.
A. Sol is the flagship for the toughest work: long-horizon coding brokers, safety analysis, and deep evaluation. Terra matches GPT-5.5 high quality at half the worth, making it the migration goal for manufacturing workloads. Luna is the quickest, most cost-effective tier but nonetheless lands close to GPT-5.5 on a number of assessments.
A. Per million tokens: Sol is $5 enter and $30 output, Terra $2.50 and $15, Luna $1 and $6. Sol Quick is a brand new premium possibility at $12.50 and $75 that serves the identical flagship mannequin at as much as 750 tokens per second on Cerebras {hardware}.
A. Sol is OpenAI’s most succesful cybersecurity mannequin, so on the authorities’s request underneath a brand new cyber Govt Order framework, the June 26 launch started as a restricted preview for roughly 20 vetted organizations. After further testing and company conferences, the Commerce Division authorized the broad launch twelve days later.
A. OpenAI classifies all three fashions at its “Excessive” cyber threat degree, with Sol fixing 96.7% of inside capture-the-flag challenges, however says none can autonomously run a whole assault marketing campaign underneath take a look at situations. They ship with 5 layered safeguards hardened by over 700,000 GPU hours of red-teaming.
I specialise in reviewing and refining AI-driven analysis, technical documentation, and content material associated to rising AI applied sciences. My expertise spans AI mannequin coaching, knowledge evaluation, and knowledge retrieval, permitting me to craft content material that’s each technically correct and accessible.
Login to proceed studying and revel in expert-curated content material.
