Radar · arena · as of September 7, 2026
The gap between open weights and the closed frontier, where the releases come from, and what inference costs. Resolution runs through public leaderboards, Hugging Face, and Epoch AI.
Sharpest call. 70% that On 2028-12-31 the top-ranked model on the Artificial Analysis Intelligence Index carries a list output price of at least $20 per million tokens on its first-party API, even as trailing-tier prices keep falling. The card.
by end-2027 · Brier-scored
60%vs 35% implied · engine 61%
On 2027-12-31 the gap between the top-scoring model and the top-scoring open-weights model on the Artificial Analysis Intelligence Index is at least 4 points on the index version then live.
clearly above consensus · resolves December 31, 2027the card ↗
by end-2029 · milestone ladder
70%vs 20% implied · engine 62%
On 2028-12-31 the top-ranked model on the Artificial Analysis Intelligence Index carries a list output price of at least $20 per million tokens on its first-party API, even as trailing-tier prices keep falling.
far above consensus · 3 rungs · resolves December 31, 2028the card ↗
50%vs 50% implied · engine 43%
On each of 2027-12-31, 2028-12-31, and 2029-12-31 the highest-scoring open-weights model on the Artificial Analysis Intelligence Index is developed by a lab headquartered in mainland China.
in line with consensus · 4 rungs · resolves December 31, 2029the card ↗
Public series with the on-track and off-track thresholds registered with each thesis. Status is computed, not narrated.
Output price of the top-ranked AA modelno data
— USD per 1M output tokens
on track >= 25 · off < 15artificial-analysis ↗
connector returned no usable reading · checked 2026-09-07
Cheapest model within 5 index points of the topno data
— USD per 1M output tokens
on track >= 5 · off < 2artificial-analysis ↗
connector returned no usable reading · checked 2026-09-07
Top open-weights Intelligence Index scoreno data
— index points
on track <= 64 · off > 67artificial-analysis ↗
connector returned no usable reading · checked 2026-09-07
Top overall Intelligence Index scoreno data
— index points
on track >= 69 · off < 67artificial-analysis ↗
connector returned no usable reading · checked 2026-09-07
Hugging Face downloads, leading Chinese open-weights orgsoff track
317.6M downloads per month
on track < 60M · off >= 100Mhf-hub ↗
series as of 2026-09-07
Developer HQ of the top open-weights AA modelno data
— boolean (1 = mainland China HQ)
on track == 1 · off == 0artificial-analysis ↗
connector returned no usable reading · checked 2026-09-07
Hugging Face downloads, Meta Llama organizationon track
25.5M downloads per month
on track < 30M · off >= 60Mhf-hub ↗
series as of 2026-09-07
Count of Chinese-lab open-weights models in the AA top 20 overallno data
— count
on track >= 3 · off <= 1artificial-analysis ↗
connector returned no usable reading · checked 2026-09-07