GPT-6 Astra is cheaper than Anthropic's best, not smarter
SentiSense Original · Live · AI-assisted, editor-reviewed
OpenAI released GPT-6 Astra on September 3 with a Brockman line about the AGI era, but the independent scoreboard split rather than flipped. Anthropic's Fable 5.1 still leads Artificial Analysis on both the Intelligence Index (66 to 61) and the Coding Agent Index (70 to 67), while Astra takes the cost line at $4.72 per task against $9.18, plus new records from Epoch AI, Perplexity and Cognition on agentic and long-horizon work. API pricing is $10 and $50 per million tokens, exactly matching Fable 5.1 and about 2.5 times OpenAI's own prior model. More third-party reviews are expected over the coming days.
OpenAI released GPT-6 Astra on September 3, calling it a generational leap in capability and, in president Greg Brockman's closing words at the press briefing, the start of "the AGI era" . The model went first to a limited set of trusted enterprise partners, with ChatGPT Plus, Pro, Business and Enterprise access following over the coming days . API list price is $10 per million input tokens and $50 per million output tokens, about 2.5 times its own predecessor GPT-5.6 Sol and an exact match for Anthropic's Fable 5.1 .
The independent scoreboard did not follow the press release. On Artificial Analysis's Intelligence Index, Astra scores 61, roughly five points behind Fable 5.1 at 66 and level with the GPT-5.6 Sol it replaces. On the Coding Agent Index it reaches 67, behind Fable 5.1 at 70 . What actually changed hands is the cost line: Artificial Analysis puts Astra at $4.72 per task against Fable 5.1's $9.18, and roughly 70% more token-efficient than Sol in the Codex harness .
That cost-per-capability read is what the application layer picked up within hours. Cognition said Astra lands within 0.4 points of Fable 5 on its FrontierCode 1.1 benchmark at 64% lower cost and is shipping into Devin immediately . Perplexity scored it 0.682 at $11.98 per task on WANDR, the highest of any model it has tested, 13.5% above Fable 5.1 at 6.1% lower cost . Epoch AI logged a new Capabilities Index record of 169, up from a prior best of 163, and the first non-zero score any model has posted on FrontierMath Erdos .
The headline numbers carry asterisks worth reading. Astra's 99.9% on ARC-AGI-3 holds only with OpenAI's own Provider Adapter harness; on the standard harness it tops out near 62.7% , and ARC Prize costed the run at roughly $360 per game . Astra is also the first OpenAI model rated "Critical" for cyber capability under the company's Preparedness Framework, so the sharpest security functionality is gated to partner organizations rather than shipped broadly . OpenAI's own claimed hallucination rate falls to 2% from 9.4% on Sol .
For the AI trade, the read is compute and distribution rather than a single vendor's leaderboard position. MSFT closed the session at $510.12, up 2.7%, and carries a Bullish SentiSense Score that is warming; NVDA at $228.45 is up 1.8% with a Strong Bullish Score and the third-highest social dominance rank on the platform; GOOGL, which both competes with OpenAI and backs Anthropic, sits at $342.48 and holds the top dominance rank. SentiSense Market Mood reads 51.2, squarely neutral and cooling on the week, which is not the tape of a market that thinks AGI arrived today. The thing to watch is not the launch post but the next several days of independent evaluation, because pricing at Fable 5.1 parity means Astra has to win on measured cost per finished task, not on the headline.
Timeline
INITIAL · Sep 3, 8:39 PM ET
OpenAI released GPT-6 Astra to a limited set of trusted enterprise partners, with ChatGPT Plus, Pro, Business and Enterprise access to follow over the coming days . API list price is $10 per million input tokens and $50 per million output tokens, matching Anthropic's Fable 5.1 . First independent reads split: Artificial Analysis has Fable 5.1 ahead on both the Intelligence Index (66 to 61) and the Coding Agent Index (70 to 67), while Astra takes cost per task at $4.72 against $9.18 . More third-party evaluations are expected through the weekend.
Powered by SentiSense - Intelligent Market Analysis