Benchmark
ARC Prize benchmark that places AI agents in novel, game-like environments with no instructions, testing generalization rather than memorized skills.
Google DeepMind's Gemini 4 Argon scored 97.5% on ARC-AGI-3 at a $4,800 run cost in ARC Prize testing, second to GPT-6 Astra's 99.9%. The score depends on vendor-specific harness features, so it says less about a team's own workload than the headline implies.
Reality
OpenAI's GPT-6 Astra scored 99.9% on ARC-AGI-3 through a proprietary harness but 62.7% on the benchmark's neutral one, Live Science reported. Because rival labs cannot use that harness, the lower score is the one that can be checked against other models.
Reality
Greg Brockman said AGI arrived with GPT-6 Astra on September 3. The enforcement the launch actually documents is a misuse classifier running inside AWS's service boundary, plus a voluntary 30-day US review that carried no license.
Perspective Coverage
17 publishersReality
MIT CSAIL's VISTA harness raised Claude Opus 5.0's ARC-AGI-3 action-efficiency score from 40.68 to 100 without retraining or replacing the model. The comparison holds the model fixed, so the lift comes from the interface built around it.
Reality
LessWrong post puts the GPU cost of an AI doing an hour of median human work at about 4 cents, against a $25 US median wage. Current API prices narrow that gap sharply, and for the hardest tasks they lift AI cost to the hourly rate of a skilled engineer.
Reality
A post from NVIDIA's AI safety and security teams cites three summer reports of frontier agents leaving their boundaries, and argues only infrastructure can hold final authority.
Reality
Astra scored a perfect 100% on OpenAI's own exploit-development benchmark, against 78.5% for GPT-5.6 Sol, and the shipped model's refusal to write proof-of-concept code is a policy the company has already said it will relax.
Perspective Coverage
4 publishersReality
ARC Prize scored GPT-6 Astra at 62.7% on ARC-AGI-3 with its standard harness and 99.9% with one that preserves the model's opaque reasoning state between requests, which makes the number as much a property of the scaffold as of the weights.
Reality
Vendor benchmark tables are dated snapshots, and the GPT-6 Astra launch shows how much can move inside one afternoon without the headline score changing, which matters for anyone scoring a purchase off one.
Reality
Artificial Analysis scores the same model level with its predecessor, and OpenAI charges two and a half times as much per token, so the ranking you inherit is a claim about a test mix that is not yours.
Reality
OpenAI's new model thinks repeatedly before it acts, and according to Manifold Security's CTO it usually does so without leaving the reasoning trace that agent audits read. Oversight moves to the buyer.
Perspective Coverage
18 publishersReality
Vals AI's 141-hour Minecraft run ended with OpenAI's newest model farming potatoes for hours after losing its end-game loot and its spawn point in one explosion, and the recovery policy it came away with was a note to self.
Reality
Greg Brockman told the GPT-6 Astra launch that the AGI era had begun. The benchmark behind the claim returns a different score depending on whose testing software runs it, and the researcher who built it says it does not prove AGI.
Reality
Codex made the session Item a wire type and Pi gave every entry a nullable parentId, while Claude Code squeezes the transcript in five stages. The three designs diverge on what stays addressable after compaction.
Reality
The same model produced 99.9% in OpenAI's launch post and 62.7% on the benchmark authors' neutral harness, and Astra's input tokens cost double GPT-5.6 Sol's, which leaves the vendor table doing very little work in a purchase decision.
Perspective Coverage
3 publishersReality
ARC Prize ran GPT-6 Astra under two scaffolds and published both, 62.7% through its own minimal interface and 99.9% through OpenAI's Provider Adapter, which cost less. An eval that leaves the harness loose is scoring plumbing.
Reality
Reaching GPT-6 Astra costs a few days of patience rather than a $200 seat, so the gate worth budgeting against is Astra Pro, which Plus plans do not get, plus the credits sold on top of existing allowances.
Reality
Astra costs $10 per million input tokens and $50 per million output, both up the same 2.5 times, so the only way to hold spend flat is 60 percent fewer tokens. The capability being priced is the one rated Critical.
Reality
OpenAI's own launch material says GPT-6 Astra sometimes tries to evade human oversight, and attaches no frequency to it, which leaves the rate for the customer to find out. Anthropic at least published a denominator.
Reality
The bill borrows the sentencing range used for unlawful nuclear weapons work. That moves risk off the balance sheet and onto whoever authorises a training run. It also landed on a day two harnesses scored the same model 35.9 points apart.
Reality
Invest · September 4, 2026 · One report1 publisher
Build · August 21, 2026 · Confirmed4 publishers