Frontier timeline · Companies

Who holds the frontier?

See which companies offer the smartest models at each budget.

Replayed using 2026-09-22 scores and prices, not historical measurements.

Company frontier share

How share works ↗

Share of the budget range where each company leads.

On this date

Latest shares

Latest shares
RankCompanyShareModelsRank change
1OpenAI48.76%2—
2Xiaomi31.52%1New
3Anthropic10.99%1—
4IBM6.01%1↑ 1
5Meta2.73%1↑ 1

Budget: $14.90–$20000 per benchmark run.

Overall leaderboard

Average share across this range

2025-09-22–2026-09-22

Overall leaderboard
RankCompanyAverage share
1OpenAI44.23%
2Anthropic30.38%
3DeepSeek10.18%
4Google4.86%
5Xiaomi2.52%
6MiniMax2.46%
7Z AI2.05%
8Mistral1.95%
9IBM0.48%
10Alibaba0.41%
11Meta0.24%
12SpaceXAI0.20%
13StepFun0.06%

How frontier share works

A company leads wherever it offers the smartest model you can afford. Its share is the slice of the budget range it covers—not market share.

  • Daily shares total 100%. Budgets start at that day’s cheapest model and end at a shared upper limit. The scale is logarithmic: each tenfold budget increase has equal width. Costs are per benchmark run, not per token.
  • Longer ranges use averages. Bars show days, weeks, or months. The leaderboard averages daily shares across the selected range. Absent companies count as zero; days with no frontier are excluded.
  • Ties split the share. Identical scores and costs split equally between companies. Duplicate models and effort variants add no extra share. For equal scores, cheaper wins.
  • Dates follow model releases. Only dated models with scores and costs are included. Open weights uses its own frontier. New snapshots can change the replay and upper budget limit.
  • Table details. Models combine effort variants. Rank change compares the previous frontier change; “New” means a company entered or returned. Periods are clipped to the selected range and percentages are rounded.

Which releases moved the frontier forward? →

View the data