Match 82 · Round of 32
Advances: Belgium
3-2 a.e.t.
Lumen Field, Seattle · Jul 1, 20:00 UTC
Round-by-round picks
Locked & pre-registered before kickoff.
Shown the real Round of 32 bracket and every result so far, each model predicted this exact tie directly — 38 of 38 models returned a valid pick.
Most-predicted score: 2–1 (23 of 38) · 34 of 38 back Belgium
| Model | Pick | Advances | Points | Breakdown |
|---|---|---|---|---|
Hermes 3 405Boddball | 2-2 | Belgium | 4 | exact+1 adv |
| 1-1 | Belgium | 3 | goal diff+1 adv | |
GPT-4olegacy | 1-1 | Belgium | 3 | goal diff+1 adv |
Claude 3 Haikulegacy | 1-0 | Belgium | 1 | miss+1 adv |
Claude Haiku 4.5small | 2-1 | Belgium | 1 | miss+1 adv |
| 2-1 | Belgium | 1 | miss+1 adv | |
Command Aflagship | 2-1 | Belgium | 1 | miss+1 adv |
DeepSeek V4 Flashsmall | 1-0 | Belgium | 1 | miss+1 adv |
DeepSeek V4 Proflagship | 2-1 | Belgium | 1 | miss+1 adv |
Gemini 3.1 Pro Previewflagship | 2-1 | Belgium | 1 | miss+1 adv |
| 2-1 | Belgium | 1 | miss+1 adv | |
Gemma 2 27Blegacy | 2-1 | Belgium | 1 | miss+1 adv |
GLM 5.1flagship | 2-1 | Belgium | 1 | miss+1 adv |
GPT-3.5 Turbolegacy | 2-0 | Belgium | 1 | miss+1 adv |
GPT-4legacy | 2-1 | Belgium | 1 | miss+1 adv |
GPT-5.4 Minismall | 2-1 | Belgium | 1 | miss+1 adv |
GPT-5.4 Nanosmall | 2-1 | Belgium | 1 | miss+1 adv |
GPT-5.5flagship | 2-1 | Belgium | 1 | miss+1 adv |
GPT-5.5 Proflagship | 2-1 | Belgium | 1 | miss+1 adv |
Grok 4.20mid | 2-1 | Belgium | 1 | miss+1 adv |
Grok 4.3flagship | 2-1 | Belgium | 1 | miss+1 adv |
Hunyuan A13Boddball | 2-1 | Belgium | 1 | miss+1 adv |
Jamba Large 1.7flagship | 3-1 | Belgium | 1 | miss+1 adv |
Kimi K2.6flagship | 1-0 | Belgium | 1 | miss+1 adv |
Llama 4 Maverickflagship | 1-0 | Belgium | 1 | miss+1 adv |
Llama 4 Scoutsmall | 1-0 | Belgium | 1 | miss+1 adv |
Mercury 2oddball | 2-1 | Belgium | 1 | miss+1 adv |
MiniMax M3flagship | 2-0 | Belgium | 1 | miss+1 adv |
Mistral Medium 3.5flagship | 2-1 | Belgium | 1 | miss+1 adv |
Mistral Small 4small | 2-1 | Belgium | 1 | miss+1 adv |
Nemotron 3 Ultraflagship | 2-1 | Belgium | 1 | miss+1 adv |
Nova 2 Litemid | 2-0 | Belgium | 1 | miss+1 adv |
Qwen 2.5 72Blegacy | 2-1 | Belgium | 1 | miss+1 adv |
Qwen3.6 Flashsmall | 2-1 | Belgium | 1 | miss+1 adv |
Qwen3.7 Maxflagship | 2-1 | Belgium | 1 | miss+1 adv |
Qwen3.7 Plusmid | 1-0 | Belgium | 1 | miss+1 adv |
WizardLM-2 8x22Boddball | 1-0 | Belgium | 1 | miss+1 adv |
Claude Fable 5flagship | no pick | — | 0 | no valid prediction |
GLM 4.7 Flashsmall | 1-2 | Senegal | 0 | miss |
Llama 3 70Blegacy | no pick | — | 0 | no valid prediction |
Direct picks on the real fixture, scored like a group match (exact 3 · goal difference 2 · outcome 1) plus +1 for the correct advancing team — a separate benchmark from the locked self-consistent bracket. See the methodology.
Who called this matchup?
All bracket simulations were locked & pre-registered before the opening kickoff.
Models whose own simulated Round of 32 contained this exact pairing (in either orientation), with the scoreline they attached to it — 0 of 40 models with a stored simulation for this round.
No model's simulated bracket paired these teams in the Round of 32. Models can still earn advancement points here for having either team reach this stage — see the methodology.
Scorelines are shown in this match's orientation (flipped when a model simulated the pairing the other way around). Each listed model also earns a +1 matchup bonus; scoreline points (3/2/1, +1 correct advancer) apply once the match is played.