Will an open-weights model be in the top 5 of the Text Arena leaderboard on 31 March 2027?
deadline 2027-03-31
confidence medium
tier 4 evidence
22%PolySignal estimate
⚠ Evidence cap applied: the model proposed 30%, brought back to 22% for lack of sources solid enough to justify the departure from the base rate.
Resolution criterion
Yes if, when the arena.ai text leaderboard is consulted on 31 March 2027, at least one of the five highest-ranked models has publicly downloadable weights. If there is a tie in score at the boundary of 5th place, all tied models are included in the check. The publisher's licence page prevails in case of doubt.
Verified starting state
The top 5 on 17 September 2026 is entirely proprietary: Claude Fable 5 (1506), Claude Opus 4.6 High (1505), Claude Opus 4.7 High (1502), Muse Spark 1.2 xHigh (1500), Claude Fable 5.1 Max (1498). The frontier/open-weight gap narrowed during 2026 without any open model breaking into the top.
What pushes it up
- The performance gap between open-weights and proprietary models narrowed during 2026, increasing the likelihood of a breakthrough [tier 1 starting state].
- Anticipated releases of next-generation open-weights models (such as Meta's Llama 4 or new Qwen/DeepSeek models) could temporarily capture top-5 spots [general knowledge].
What pushes it down
- The top 5 is currently entirely proprietary, dominated by advanced models like Claude Fable 5 and Claude Opus 4.7 [tier 1 starting state].
- Proprietary labs iterate rapidly and can deploy larger, more resource-intensive models that are difficult to release as open-weights, quickly displacing any open-weights challengers [general knowledge].
What would move this number most
The release timing and performance of Meta's Llama 4 relative to upcoming proprietary models from Anthropic and OpenAI.
How this number is built
| Deadline | 2027-03-31 |
| Base rate | 15% |
| Best evidence | tier 4 · 0 article(s) used |
| Independent runs | 20 · 23 (spread 3 pts) |
| Confidence | medium |
| Evidence cap | applied: 30% brought back to 22% |
| Last revised | 2026-09-23 09:04 |
| Model | gemini-3.5-flash |