The Token Tape

daily brief ·

Codebuff overtakes Codex in OpenRouter usage; Claude Code Pro jumps $17→$200/yr

Usage Codebuff passed Codex on OpenRouter coding tokens, 272,013,206,554 vs 251,965,943,112 over 7 days (+8% margin), even as both fell from prior-week levels of 332,823,865,013 and 344,392,183,290 respectively (evt 45190).

Plans Claude Code overhauled its tiers: Free now lists $0/yr, Pro jumps from $17→$200/yr, and Team seat plans were restructured into "Team - Standard seat" and "Team - Premium seat" (evt 45175). Zencoder, Trae, Roo Code, Qodo, Lovable, Kilo Code, Junie, Firebase Studio, Replit, and Emergent all shipped smaller plan/limit adjustments today (evts 45189, 45188, 45187, 45186, 45185, 45184, 45183, 45182, 45181, 45180).

Prices Mixed moves across models. Ling-3.0-flash saw input cut 72% to $0.021/MTok on OpenRouter and models.dev (evt 44967, 44969), while its kilo listing fell 20% to $0.06 (evt 44968); nano-gpt versions rose 25% to $0.075 (evt 44961, 44962). Qwen3-vl-235b-a22b-thinking input dropped 59.2% to $0.4/MTok (evt 44793, 44419). Nvidia nemotron-3-super-120b-a12b input jumped 252.9% to $0.3/MTok (evt 44762, 44403). Qwen3.6-27b input rose 107.6% to $0.6/MTok (evt 42432). GLM-5.2 input eased 8.8% to $0.693/MTok (evt 44367). Cortecs GLM-4.7 input rose 73.3% to $0.78/MTok (evt 42696).

Benchmarks Anthropic's claude-opus-5:batch scored 78 on aa-coding-index, 63.1 on aa-intelligence-index, and 59.2 on aa-agentic-index (evt 46263, 46262, 46264). OpenAI's gpt-5.6-sol:batch and gpt-5.6-terra:batch posted top-5 coding-index scores of 77.4 and 76.7 respectively (evt 46276, 46273).

events in this brief

On the tape

40 detected events · each links its primary source