A $2 model ranks second on Code Arena

Qwen3.8-Max-0902 sits second on the Code Arena coding leaderboard run by Arena with a rating of 1688, one point above the third-placed Anthropic entry and 77 behind the leader. It is listed at $2 per million input tokens and $6 output, against $10 and $50 for the model in first place.

At a glance

• The board ranks 124 entries by rating, each with a confidence band, and lists input and output prices per million tokens alongside vote counts and context windows for most rows.
• The leader is an Anthropic model rated 1765 with a band of 23 points on 1,106 votes, the smallest vote total among the top six listings on the board.
• Anthropic is named on 18 of the 124 rows, including ranks one, three and six, and its entries range from $1 to $15 per million input tokens.

A second-place rating paired with $2 input pricing is the part that stands out: the leader lists $10 input against $2, and $50 output against $6. The one-point gap to third reads as noise rather than a ranking, given a confidence band of 18 points on the second entry and 8 on the third. Vote totals across the top differ by an order of magnitude, which likely matters for how settled those positions are.

Qwen3.8-Max-0902 records 1688 with a band of 18 points either way on 1,472 votes, and the board credits it with beating Claude Opus 5 max. Among the top six rows, only its listing and the fifth-placed entry, rated 1669, carry $2 input and $6 output pricing.

Third is an Anthropic entry at 1687 with a band of eight points on 10,583 votes, priced at $5 input and $25 output. Fourth is rated 1674 at $3 and $15. Sixth is another Anthropic model at 1661 on 10,530 votes with the same $5 and $25 pricing as third.

Below the top six the table runs to 124 rows, ending at a rating of 1080 with a band of 31 points on 831 votes. Named vendors further down include Tencent at rank nine with 1626 and $0.83/$2.50 pricing, Meta at rank 25, Bytedance at rank 32 and Thinking Machines at ranks 67 and 70.

Prices span $0.03 per million input tokens at rank 84, rated 1371, up to $15 input and $75 output at rank 77, an Anthropic entry rated 1389. Vote totals run from 228 on the rank-53 listing, rated 1449, to 27,470 on an Anthropic entry at rank 100 rated 1329.

What the Qwen row omits

The Qwen3.8-Max-0902 row gives no context window, listing N/A where the entries at ranks one, three, four, five and six all show 1M. The board carries no release or availability details, no evaluation dates and no breakdown of how the 1,472 votes behind the second-place rating were collected. With a confidence band of 18 points, the position could move as more votes land.