Astra Hits the Text Board. The Two-Week Clock Is Still Ticking
Astra is on the Sep 13 text board at 1480 Elo. helloai still waits on two weeks and a thicker sample before it can take a slot.
The listing last week's article said was missing is now on the board. Arena.ai's September 13, 2026 text-overall snapshot puts gpt-6-astra-max at 1480±12 Elo on 2,693 votes, rank 24. OpenAI shipped GPT-6 Astra on September 3, so this is a ten-day sample, not two weeks of sustained Elo. helloai's admission rule cares about that distinction more than the rank.
The invoice did not change while the Elo arrived. The API id is gpt-6-astra at $10 input / $50 output per million tokens, cache reads at $1, and a 1,050,000-token context. Prompts over 272K input double input and cache rates and multiply output by 1.5; Fast mode is 2x Standard. Claude Fable 5.1 is the same $10/$50 list with cache reads at $0.25 — four times cheaper than Astra's $1 cache read. A same-sticker competitor with a worse cache still has to clear the Elo gate, not the price sheet.
Code Arena is not the gate, and it already crowned Astra. The September 11, 2026 WebDev board lists gpt-6-astra-max at 1800 on 2,281 votes, rank 1, with claude-fable-5.1-max at 1758 on 3,036 votes. On text overall, Claude Fable 5.1-max is 1498±8 on 5,783 votes, and Muse Spark 1.3-max is now on that board at 1493±9 on 4,723 votes. Gemini 3.1 Pro Preview remains 1487; Qwen3.8-Max is 1481. Astra's 1480 sits a point under Qwen, with a ±12 interval wider than Fable's ±8 and Muse's ±9.
helloai tracks six frontier APIs, and the floor is now Grok 4.6 at 1456 on 15,521 votes, no longer Preliminary. Admission requires LMArena text-overall Elo within 25 points of that floor, sustained two weeks, not a fresh debut with few votes. 1480 clears the 25-point bar versus 1456; 2,693 votes and a ±12 interval on a September 13 snapshot do not clear the two-week / thin-sample bar. Admitting Astra as a new provider would drop Grok — the owner already held GLM-5.3-max at 1483 on 10,960 votes rather than make that cut.
Ten days after ship, the two-week clock is still ticking. The interesting fight is no longer whether GPT-6 Astra has a text Elo. It is whether a 2,693-vote debut at rank 24 is allowed to displace a 15,521-vote floor the table already refused to drop for GLM-5.3-max. If later snapshots keep Astra inside 25 points of Grok 4.6 once that clock runs out, the six-model rule has to name who leaves.