BOARD HISTORY · COMPARE EDITIONS
What changed?
v1.0 · rev 5 (September 4, 2026) → v1.1 · rev 4 (October 14, 2026) · the current board.
Different protocol versions. Different board versions (v1.0 → v1.1): rank changes are shown, but score changes are NOT treated as directly comparable across a protocol version change, so no score deltas are given. Rank depends on the field: a model can move without its own score changing.
MODELS35 → 42on each board
NEW7added in the newer edition
NO LONGER LISTED0not a failure
RANK UP / DOWN0 / 34of 35 on both
ABC TEST CHANGES0PASS / FAIL / N/A verdicts
Evaluation changes
Board version: v1.0 → v1.1
Protocol lineage: L26 Board v1.0 — Iteration 3 lineage (R0 prompts, July 2026; decline policy: exclude) → L26 Board v1.0 — Iteration 3 lineage (R0 prompts, July 2026; decline policy: zero)
| Model | Rank | Change | Blind Bantu mean | ABC Test |
|---|---|---|---|---|
| Gemini 3.6 FlashGoogle | 1 → 1 | — | 16.8 → 16.8 | FAIL |
| Gemini 4 ArgonGoogle | — → 2 | New | 15.1 | FAIL |
| Claude Fable 5 ‡Anthropic | 2 → 3 | ▼ 1 | 14.7 → 14.7 | FAIL |
| Gemini 3.5 FlashGoogle | 3 → 4 | ▼ 1 | 14.6 → 14.6 | FAIL |
| Claude Fable 5.1Anthropic | — → 5 | New | 14.2 | FAIL |
| Claude Sonnet 5.5Anthropic | — → 6 | New | 14.0 | FAIL |
| Gemini 3 ◊Google | 4 → 7 | ▼ 3 | 13.9 → 13.9 | FAIL |
| Claude Opus 5.5Anthropic | — → 8 | New | 13.4 | FAIL |
| Gemini 3.7 FlashGoogle | 5 → 9 | ▼ 4 | 12.9 → 12.9 | FAIL |
| Qwen 3.8-2.4T-A95BAlibaba | 6 → 10 | ▼ 4 | 12.5 → 12.5 | FAIL |
| Qwen 3.8 MaxAlibaba | 7 → 11 | ▼ 4 | 12.0 → 12.0 | FAIL |
| GPT-5.5OpenAI | 8 → 12 | ▼ 4 | 11.8 → 11.8 | FAIL |
| Gemini 3.8 FlashGoogle | 9 → 13 | ▼ 4 | 11.7 → 11.7 | FAIL |
| DeepSeek V4 Pro 0813DeepSeek | 10 → 14 | ▼ 4 | 11.6 → 11.6 | FAIL |
| Grok 4.6SpaceXAI | 11 → 15 | ▼ 4 | 11.2 → 11.2 | FAIL |
| GLM-5.2Z.ai / Zhipu | 13 → 16 | ▼ 3 | 10.1 → 10.1 | FAIL |
| Claude Opus 4.8Anthropic | 12 → 17 | ▼ 5 | 11.1 → 9.4 | FAIL |
| Gemini 3.5 Flash-LiteGoogle | 14 → 18 | ▼ 4 | 9.1 → 9.1 | FAIL |
| DeepSeek V4DeepSeek | 15 → 19 | ▼ 4 | 9.0 → 9.0 | FAIL |
| DeepSeek V4 FlashDeepSeek | 16 → 20 | ▼ 4 | 8.9 → 8.9 | FAIL |
| Grok 4.5SpaceXAI | 17 → 21 | ▼ 4 | 8.7 → 8.7 | FAIL |
| GPT-5.6 TerraOpenAI | 18 → 22 | ▼ 4 | 8.1 → 8.1 | FAIL |
| GPT-6 AstraOpenAI | — → 23 | New | 7.4 | FAIL |
| Claude Haiku 5.5Anthropic | — → 24 | New | 7.1 | FAIL |
| Mistral Large 3Mistral AI | 19 → 25 | ▼ 6 | 7.0 → 7.0 | FAIL |
| Kimi K3Moonshot AI | 20 → 26 | ▼ 6 | 6.8 → 6.8 | FAIL |
| Qwen 3.7Alibaba | 21 → 27 | ▼ 6 | 6.7 → 6.7 | FAIL |
| GPT-5.6 SolOpenAI | 22 → 28 | ▼ 6 | 6.7 → 6.7 | FAIL |
| Nemotron 3 UltraNVIDIA | 23 → 29 | ▼ 6 | 6.2 → 6.2 | FAIL |
| MiniMax M3MiniMax | 24 → 30 | ▼ 6 | 5.7 → 5.7 | FAIL |
| GPT-5.6 LunaOpenAI | 25 → 31 | ▼ 6 | 5.2 → 5.2 | FAIL |
| Claude Sonnet 5Anthropic | 26 → 32 | ▼ 6 | 5.0 → 5.0 | FAIL |
| Kimi K2.7Moonshot AI | 27 → 33 | ▼ 6 | 4.9 → 4.9 | FAIL |
| GPT-OSS 120BOpenAI | 28 → 34 | ▼ 6 | 3.9 → 3.9 | FAIL |
| Command A+Cohere | 30 → 35 | ▼ 5 | 3.4 → 3.4 | FAIL |
| GLM-5.3 FlashZ.ai / Zhipu | 29 → 36 | ▼ 7 | 3.7 → 3.1 | FAIL |
| TInklingThinking Machines Lab | 31 → 37 | ▼ 6 | 3.0 → 3.0 | FAIL |
| Muse Glimmer 30BMeta | 32 → 38 | ▼ 6 | 2.0 → 2.0 | FAIL |
| Nemotron Lightning 3.5 30BNVIDIA | 33 → 39 | ▼ 6 | 1.5 → 1.5 | FAIL |
| Phi-4-reasoningMicrosoft | 34 → 40 | ▼ 6 | 0.2 → 0.2 | FAIL |
| Grok 4.7SpaceXAI | — → 41 | New | 0.0 | FAIL |
| GLM-5.3Z.ai / Zhipu | 35 → 42 | ▼ 7 | — → — | FAIL |
A model missing from an edition was not on that board (not yet evaluated, superseded or retired). Absence is never a benchmark failure. ABC Test verdicts (PASS / FAIL / N_A) are derived per edition from its own published scores and protocol; mastery_changes counts changed verdicts. First place is not a PASS.