Opus 5.5 Cuts the Price. Grok 4.7 Keeps Its Rate
Claude Opus 5.5 drops to $4/$20 while Grok 4.7 keeps $2/$6. Text Arena is still the September 13 snapshot, so both cards keep the prior Elo.
Anthropic cut the Opus list price on September 22, 2026, and helloai is replacing Claude Opus 5 with Claude Opus 5.5 in the same row. The API id is claude-opus-5-5. Input falls from $5 to $4 per million tokens, output from $25 to $20, and cache reads from $0.50 to $0.20. Anthropic says the model performs at the level of Claude Fable 5.1 on most work, costs about 40% less to run than Opus 5 on typical workloads, and generates output more than 30% faster. The context window stays 1 million tokens.
The invoice moved before the leaderboard did. Arena.ai's text-overall table is still the September 13 snapshot, nine days before Claude Opus 5.5 shipped, and the September 22 Code Arena WebDev board still lists claude-opus-5-max and claude-opus-5-high rather than a 5.5 slug. helloai keeps the prior text score on the card: claude-opus-5-high at 1493. New aliases prefer claude-opus-5.5 names and fall back to the Opus 5 slugs, so the next live scrape can replace that number without inventing one today.
xAI's release is a version bump with the same bill. Grok 4.7 shipped on September 21, and the developer docs now say to use Grok 4.7 for chat and for code. The API id is grok-4.7. The window is still 500,000 tokens. Standard rates are still $2 input and $6 output per million tokens when the prompt stays under 200,000 tokens, and $4 and $12 once a prompt crosses that line. SpaceXAI's launch post says the model is served at the same price and speed as Grok 4.6.
The only independent score for Grok 4.7 is a one-day code sample. On the September 22 WebDev board, grok-4.7-xhigh sits at 1632 with a plus-or-minus 17 interval on 1,425 votes, rank 10. grok-4.6-high is 1616 on 6,447 votes. gpt-6-astra-max still leads that board at 1793 on 4,230 votes, and claude-fable-5.1-max is 1755 on 4,887. Text overall has not listed grok-4.7. The card Elo stays 1456, which is grok-4.6-high on the September 13 text snapshot, not a measurement of 4.7.
The six-model set does not grow on either launch. GPT-6 Astra's only text score is still 1480 plus-or-minus 12 on 2,693 votes from September 13, ten days after its September 3 debut. The model is twenty days old today, but the board has not published a second point, so the two-week sustainment bar is still unmet. Admitting OpenAI would mean dropping a tracked row at a cap of six. helloai updates Claude Opus 5.5 and Grok 4.7 because the catalogs did, and it leaves both Elo fields on the last slug the text board actually scored.