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%Asturna estimate
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 gap between frontier proprietary and open-weights models narrowed during 2026, increasing the potential for an upcoming open flagship to contend for the top tier [tier 1].
- Leaderboard rankings can fluctuate as new evaluation sets and prompt distributions are introduced, occasionally allowing competitive new releases to peak high [tier 4, article 1].
What pushes it down
- As of 17 September 2026, the top 5 is entirely proprietary models clustered tightly above 1498 Elo, with four spots occupied by Anthropic variants alone [tier 1].
- No open-weight model succeeded in breaking into the top tier during 2026 despite the narrowing gap [tier 1].
- With only about 5.5 months until 31 March 2027, proprietary developers are likely to continue deploying iterative updates that crowd the top 5 [tier 1].
What would move this number most
Whether an open-weight developer releases a next-generation flagship model before March 2027 that surpasses existing Claude and Muse frontier checkpoints without proprietary models immediately leapfrogging it.
How this number is built
| Deadline | 2027-03-31 |
| Base rate | 18% |
| Best evidence | tier 4 · 1 article(s) used |
| Independent runs | 20 · 24 (spread 4 pts) |
| Confidence | medium |
| Evidence cap | not triggered |
| Last revised | 2026-10-10 12:43 |
| Model | gemini-3.8-flash |
Sources consulted
Probability history
2026-10-10 12:43 22% 5 article(s) on file · gemini-3.8-flash
- What pushes it up. The gap between frontier proprietary and open-weights models narrowed during 2026, increasing the potential for an upcoming open flagship to contend for the top tier [tier 1].
- What pushes it up. Leaderboard rankings can fluctuate as new evaluation sets and prompt distributions are introduced, occasionally allowing competitive new releases to peak high [tier 4, article 1].
- What pushes it down. As of 17 September 2026, the top 5 is entirely proprietary models clustered tightly above 1498 Elo, with four spots occupied by Anthropic variants alone [tier 1].
- What pushes it down. No open-weight model succeeded in breaking into the top tier during 2026 despite the narrowing gap [tier 1].
- What pushes it down. With only about 5.5 months until 31 March 2027, proprietary developers are likely to continue deploying iterative updates that crowd the top 5 [tier 1].
- What would move this number most. Whether an open-weight developer releases a next-generation flagship model before March 2027 that surpasses existing Claude and Muse frontier checkpoints without proprietary models immediately leapfrogging it.
2026-10-07 09:18 6% 3 article(s) on file · openai/gpt-oss-20b (Groq)
- What would move this number most. Whether an open‑weight model will break into the top 5 by March 2027, which depends on future model releases and leaderboard dynamics.
2026-09-23 09:04 22% 2 article(s) on file · gemini-3.5-flash
Reasons not recorded for this revision (detailed history has only been kept since recently).