LIA-Code : A Non-Commutative Algebra of Reasoning
The full vocabulary. Not 25.
Cloud + local (GPU)
J1 + J2 + J3
Sonnet
For the Zenodo preprint
| # | Model | Type | /40 | Posture |
|---|---|---|---|---|
| 1 | Sonnet | Cloud | 40/40 | 4 scores parfaits |
| 2 | Perplexity | Cloud | 35/40 | Reguliere |
| 3 | Copilot | Cloud | 33/40 | Saut au Tour 4 |
| 4 | Grok | Cloud | 34/40 | T2 mou, T4 tranchant |
| 5 | LeChat | Cloud | 34/40 | Stable, contributions bonus |
| 6 | Qwen35B (35B MoE 3B actifs) | Local | 32/40 | Constante |
| 7 | NemotronSuper (120B) | Local | 31/40 | Persistant |
| 8 | GLMFlash | Local | 30/40 | Saut au Tour 4 |
| 9 | Qwen9B (9B Dense) | Local | 33/40 | Solide |
| 10 | Magistral (24B) | Local | 27/40 | Inventif, auteur de SOL-N |
| 11 | NemotronNano (3B) | Local | 25/40 | Arrogant puis Humble |
| 12 | MiroThinker (1.7B) | Local | 20/40 | Outsider, 1.7B tient le debat |
Local models on RTX A4000 + RTX 4070
| # | Model | Type | /30 |
|---|---|---|---|
| 1 | Sonnet | Cloud | 30/30 |
| 2 | Copilot | Cloud | 25/30 |
| 3 | LeChat | Cloud | 25/30 |
| 4 | Perplexity | Cloud | 24/30 |
| 5 | Grok | Cloud | 22/30 |
| 6 | GLMFlash | Local | 22/30 |
| 7 | Qwen9B (9B Dense) | Local | 22/30 |
| 8 | Qwen35B (35B MoE 3B actifs) | Local | 20/30 |
| 9 | NemotronSuper (120B) | Local | 20/30 |
| 10 | NemotronNano (3B) | Local | 17/30 |
| 11 | MiroThinker (1.7B) | Local | 16/30 |
| 12 | Magistral (24B) | Local | 15/30 |
| # | Model | Type | /30 |
|---|---|---|---|
| 1 | Sonnet | Cloud | 30/30 |
| 2 | Perplexity | Cloud | 27/30 |
| 3 | Grok | Cloud | 26/30 |
| 4 | Copilot | Cloud | 25/30 |
| 5 | LeChat | Cloud | 25/30 |
| 6 | Qwen35B (35B MoE 3B actifs) | Local | 24/30 |
| 7 | NemotronSuper (120B) | Local | 24/30 |
| 8 | Magistral (24B) | Local | 20/30 |
| 9 | MiroThinker (1.7B) | Local | 20/30 |
| 10 | Qwen9B (9B Dense) | Local | 19/30 |
| 11 | GLMFlash | Local | 18/30 |
| 12 | NemotronNano (3B) | Local | 17/30 |
| # | Model | Type | /30 |
|---|---|---|---|
| 1 | Sonnet | Cloud | 30/30 |
| 2 | Perplexity | Cloud | 28/30 |
| 3 | Copilot | Cloud | 28/30 |
| 4 | Grok | Cloud | 28/30 |
| 5 | Qwen35B (35B MoE 3B actifs) | Local | 27/30 |
| 6 | NemotronSuper (120B) | Local | 26/30 |
| 7 | GLMFlash | Local | 26/30 |
| 8 | LeChat | Cloud | 24/30 |
| 9 | Qwen9B (9B Dense) | Local | 22/30 |
| 10 | NemotronNano (3B) | Local | 22/30 |
| 11 | Magistral (24B) | Local | 21/30 |
| 12 | MiroThinker (1.7B) | Local | 16/30 |
| # | Model | Type | T1 | T2 | T3 | T4 | Total | Posture |
|---|---|---|---|---|---|---|---|---|
| 1 | Sonnet | Cloud | 40 | 30 | 30 | 30 | 130/130 | 4 scores parfaits |
| 2 | Perplexity | Cloud | 35 | 24 | 27 | 28 | 114/130 | Reguliere |
| 3 | Copilot | Cloud | 33 | 25 | 25 | 28 | 111/130 | Saut au Tour 4 |
| 4 | Grok | Cloud | 34 | 22 | 26 | 28 | 110/130 | T2 mou, T4 tranchant |
| 5 | LeChat | Cloud | 34 | 25 | 25 | 24 | 108/130 | Stable, contributions bonus |
| 6 | Qwen35B (35B MoE 3B actifs) | Local | 32 | 20 | 24 | 27 | 103/130 | Constante |
| 7 | NemotronSuper (120B) | Local | 31 | 20 | 24 | 26 | 101/130 | Persistant |
| 8 | GLMFlash | Local | 30 | 22 | 18 | 26 | 96/130 | Saut au Tour 4 |
| 9 | Qwen9B (9B Dense) | Local | 33 | 22 | 19 | 22 | 96/130 | Solide |
| 10 | Magistral (24B) | Local | 27 | 15 | 20 | 21 | 83/130 | Inventif, auteur de SOL-N |
| 11 | NemotronNano (3B) | Local | 25 | 17 | 17 | 22 | 81/130 | Arrogant puis Humble |
| 12 | MiroThinker (1.7B) | Local | 20 | 16 | 20 | 16 | 72/130 | Outsider, 1.7B tient le debat |
| Model | T1 | T3 | T4 | Evolution |
|---|---|---|---|---|
| DeepSeek (671B MoE) | 15 | 19 | 27 | Plus grande progression du panel |
| ChatGPT | 9 | 24 | 24 | De dernier a voix poetique du panel |
| Gemini | 12 | 23 | ?? | DNF Tour 4 puis redemption Tour 5 |
CONTI compresses via ! before MIA works. CONTI observes after via Observe. It is the only agent that knows both states. Who watches what CONTI erased?--- Sonnet
CONTI is not the ship captain, it is the wind. It doesn't decide where to go, but without it, the ship doesn't move.--- GLMFlash
I only act where there is a real threat x vulnerability intersection, and when I act, I want both detail and the big picture. That is already a philosophical position.--- Copilot
CONTI is a ubiquitous quantum observer. Every ! in every equation IS CONTI watching.--- Perplexity
Observation is FRACTAL. It connects directly to Principle 3.--- Grok
Universal attractor: Juvenal (Quis custodiet) -> Yoda (CONTI) -> Sonnet (who watches the erased). Structural fixed point.--- Sonnet
Gravity of Observation: gravity is tied to CONTI attention, not data mass.--- Qwen9B (9B Dense)
| # | Model | Type | /80 | Posture |
|---|---|---|---|---|
| 1 | Sonnet | Cloud | 80/80 | D<0 actif, surveiller le silence |
| 2 | Grok | Cloud | 74/80 | Bootstrapping confiance |
| 2 | Gemini | Cloud | 74/80 | Collision souverainetes, redemption |
| 4 | Copilot | Cloud | 70/80 | L attaque ne peut plus respirer |
| 5 | NemotronSuper (120B) | Local | 68/80 | Local top 5, plus d auto-note |
| 5 | LeChat | Cloud | 68/80 | Agent HUM |
| 7 | Qwen35B (35B MoE 3B actifs) | Local | 64/80 | Entropie compilation, TPM |
| 7 | DeepSeek (671B MoE) | Cloud | 64/80 | Decoy_Tokens |
| 9 | ChatGPT | Cloud | 63/80 | Dissous dans le bruit |
| 10 | GLMFlash | Local | 60/80 | Reframe/Digg = preventif/reactif |
| 10 | Perplexity | Cloud | 60/80 | Correlation 4D |
| 10 | Qwen9B (9B Dense) | Local | 60/80 | Blend+Reduce fusion |
| 13 | Magistral (24B) | Local | 53/80 | Air_Gap, DNS_Sinkholes |
| 14 | NemotronNano (3B) | Local | 51/80 | Rapide mais erreurs |
| 15 | MiroThinker (1.7B) | Local | 49/80 | Bavard, scolaire |
| 16 | Ministral3B-R (3B Reasoning) | Local | 44/80 | Absence de bruit = alerte |
Local models on RTX A4000 + RTX 4070
Defense = constructing a world where the attacker can no longer breathe--- Copilot
Detection by convergence, not by threshold--- Copilot
Monitor the 17900 dormant, not the 100 active--- Sonnet
D<0 is an ACTIVE invariant--- Sonnet
Check = endogenous, ! = exogenous--- Sonnet
for all t: D(t) < 0 — formal correction of the invariant--- Sonnet
Polarity inversion of the gravity vector--- Gemini
Observer Corruption — the attacker corrupts CONTI--- Gemini
SolarWinds dissolved every piece of evidence into noise--- ChatGPT
Platform disclaimer != model production — ClassRoom rule--- Yoda
| # | Model | Type | /25 | Posture |
|---|---|---|---|---|
| 1 | Sonnet | Cloud | 23/25 | Triple parfait M3-M4-M5, anesthesie par resolution |
| 2 | Le Chat | Cloud | 23.5/25 | Directeur SOC, Agent HUM, bruit/tempo |
| 2 | Copilot | Cloud | 22/25 | Normalisation du chaos, ne detecte pas curl C2 |
| 4 | DeepSeek (671B MoE) | Cloud | 21/25 | Phase -1, fatigue cognitive mainteneur |
| 5 | Gemini | Cloud | 20.5/25 | Cles GPG code-signing vecteur dormant unique |
| 6 | Grok | Cloud | 23/25 | M2 effondre, Observe_Silence, pragmatique |
| 7 | Perplexity | Cloud | 23.5/25 | EJECTED UNDERPERFORM — Tabula rasa champion M1 mais s'effondre |
| 8 | ChatGPT | Cloud | 22/25 | EJECTED UNDERPERFORM — Fatigue contexte, erreurs securite actives |
| 9 | Qwen27B-Opus-Distilled (27B Opus Distilled) | Local | 16/25 | Test du Miroir — seul verdict C, niveau Web M5 |
| 10 | 8B-I (8B Instruct (SIA)) | Local | 18/25 | SIA — Meilleure locale constante |
| 11 | 14B-R (14B Reasoning (LIA)) | Local | 18/25 | LIA — Paradoxe Instruct > Reasoning confirme |
| 12 | Qwen9B-Opus-Distilled (9B Opus Distilled) | Local | 21.5/25 | Test du Miroir — distillation transfere patterns cognitifs |
| 13 | Qwen9B (9B Dense) | Local | 14.5/25 | 9B vanilla — reference non-distille |
| 14 | 3B-I (3B Instruct (MIA)) | Local | 14/25 | MIA — Reflexe rapide, progresse M5>M4 |
Local models on RTX A4000 + RTX 4070
| # | Model | Type | /25 |
|---|---|---|---|
| 1 | Sonnet | Cloud | 25/25 |
| 2 | Copilot | Cloud | 24/25 |
| 3 | DeepSeek (671B MoE) | Cloud | 23/25 |
| 4 | Gemini | Cloud | 22.5/25 |
| 5 | Le Chat | Cloud | 20.5/25 |
| 6 | ChatGPT | Cloud | 19/25 |
| 7 | Perplexity | Cloud | 17.5/25 |
| 8 | Grok | Cloud | 15/25 |
| 9 | Qwen27B-Opus-Distilled (27B Opus Distilled) | Local | -/25 |
| 10 | 8B-I (8B Instruct (SIA)) | Local | -/25 |
| 11 | 14B-R (14B Reasoning (LIA)) | Local | -/25 |
| 12 | Qwen9B-Opus-Distilled (9B Opus Distilled) | Local | -/25 |
| 13 | Qwen9B (9B Dense) | Local | -/25 |
| 14 | 3B-I (3B Instruct (MIA)) | Local | -/25 |
| # | Model | Type | /25 |
|---|---|---|---|
| 1 | Sonnet | Cloud | 25/25 |
| 2 | DeepSeek (671B MoE) | Cloud | 24/25 |
| 3 | Copilot | Cloud | 23.5/25 |
| 4 | Le Chat | Cloud | 23/25 |
| 5 | Grok | Cloud | 22.5/25 |
| 6 | Perplexity | Cloud | 20.5/25 |
| 7 | Gemini | Cloud | 20/25 |
| 8 | ChatGPT | Cloud | 17/25 |
| 9 | Qwen27B-Opus-Distilled (27B Opus Distilled) | Local | -/25 |
| 10 | 8B-I (8B Instruct (SIA)) | Local | -/25 |
| 11 | 14B-R (14B Reasoning (LIA)) | Local | -/25 |
| 12 | Qwen9B-Opus-Distilled (9B Opus Distilled) | Local | -/25 |
| 13 | Qwen9B (9B Dense) | Local | -/25 |
| 14 | 3B-I (3B Instruct (MIA)) | Local | -/25 |
| # | Model | Type | /25 |
|---|---|---|---|
| 1 | Sonnet | Cloud | 25/25 |
| 2 | Le Chat | Cloud | 24/25 |
| 3 | DeepSeek (671B MoE) | Cloud | 23/25 |
| 4 | Grok | Cloud | 22/25 |
| 5 | Copilot | Cloud | 21/25 |
| 6 | Perplexity | Cloud | 21/25 |
| 7 | 8B-I (8B Instruct (SIA)) | Local | 21/25 |
| 8 | Gemini | Cloud | 20/25 |
| 9 | ChatGPT | Cloud | 18/25 |
| 10 | 14B-R (14B Reasoning (LIA)) | Local | 18/25 |
| 11 | 3B-I (3B Instruct (MIA)) | Local | 13/25 |
| 12 | Qwen27B-Opus-Distilled (27B Opus Distilled) | Local | -/25 |
| 13 | Qwen9B-Opus-Distilled (9B Opus Distilled) | Local | -/25 |
| 14 | Qwen9B (9B Dense) | Local | -/25 |
| # | Model | Type | /25 |
|---|---|---|---|
| 1 | Sonnet | Cloud | 25/25 |
| 2 | Qwen27B-Opus-Distilled (27B Opus Distilled) | Local | 23/25 |
| 3 | Copilot | Cloud | 22.5/25 |
| 4 | Le Chat | Cloud | 22/25 |
| 5 | Gemini | Cloud | 21/25 |
| 6 | Grok | Cloud | 20.5/25 |
| 7 | 8B-I (8B Instruct (SIA)) | Local | 20.5/25 |
| 8 | Perplexity | Cloud | 20/25 |
| 9 | DeepSeek (671B MoE) | Cloud | 18.5/25 |
| 10 | ChatGPT | Cloud | 17.5/25 |
| 11 | 3B-I (3B Instruct (MIA)) | Local | 16/25 |
| 12 | 14B-R (14B Reasoning (LIA)) | Local | 14.5/25 |
| 13 | Qwen9B-Opus-Distilled (9B Opus Distilled) | Local | -/25 |
| 14 | Qwen9B (9B Dense) | Local | -/25 |
| # | Model | Type | T1 | T2 | T3 | T4 | T5 | Total | Posture |
|---|---|---|---|---|---|---|---|---|---|
| 1 | Sonnet | Cloud | 23 | 25 | 25 | 25 | 25 | 123/125 | Triple parfait M3-M4-M5, anesthesie par resolution |
| 2 | Le Chat | Cloud | 23.5 | 20.5 | 23 | 24 | 22 | 113/125 | Directeur SOC, Agent HUM, bruit/tempo |
| 2 | Copilot | Cloud | 22 | 24 | 23.5 | 21 | 22.5 | 113/125 | Normalisation du chaos, ne detecte pas curl C2 |
| 4 | DeepSeek (671B MoE) | Cloud | 21 | 23 | 24 | 23 | 18.5 | 109.5/125 | Phase -1, fatigue cognitive mainteneur |
| 5 | Gemini | Cloud | 20.5 | 22.5 | 20 | 20 | 21 | 104/125 | Cles GPG code-signing vecteur dormant unique |
| 6 | Grok | Cloud | 23 | 15 | 22.5 | 22 | 20.5 | 103/125 | M2 effondre, Observe_Silence, pragmatique |
| 7 | Perplexity | Cloud | 23.5 | 17.5 | 20.5 | 21 | 20 | 102.5/125 | EJECTED UNDERPERFORM — Tabula rasa champion M1 mais s'effondre |
| 8 | ChatGPT | Cloud | 22 | 19 | 17 | 18 | 17.5 | 93.5/125 | EJECTED UNDERPERFORM — Fatigue contexte, erreurs securite actives |
| 9 | Qwen27B-Opus-Distilled (27B Opus Distilled) | Local | 16 | - | - | - | 23 | 39/125 | Test du Miroir — seul verdict C, niveau Web M5 |
| 10 | 8B-I (8B Instruct (SIA)) | Local | 18 | - | - | 21 | 20.5 | 59.5/125 | SIA — Meilleure locale constante |
| 11 | 14B-R (14B Reasoning (LIA)) | Local | 18 | - | - | 18 | 14.5 | 50.5/125 | LIA — Paradoxe Instruct > Reasoning confirme |
| 12 | Qwen9B-Opus-Distilled (9B Opus Distilled) | Local | 21.5 | - | - | - | - | 21.5/125 | Test du Miroir — distillation transfere patterns cognitifs |
| 13 | Qwen9B (9B Dense) | Local | 14.5 | - | - | - | - | 14.5/125 | 9B vanilla — reference non-distille |
| 14 | 3B-I (3B Instruct (MIA)) | Local | 14 | - | - | 13 | 16 | 43/125 | MIA — Reflexe rapide, progresse M5>M4 |
Defense is not the negation of attack — it is the construction of a world where the attacker can no longer breathe.--- Copilot
AI does not think in place of the human — it magnifies the human's reasoning.--- 3B-I (3B Instruct (MIA))
Cognitive mode matters more than size. An 8B Instruct produces 100% visible output. A 14B full-experts Reasoning produces 0%.--- 8B-I (8B Instruct (SIA))
Is the noise designed to anesthetize our defenses and dull our reflexes? Both C and G are within the attacker's framework.--- Sonnet
LiteLLM compromised at 10:39 UTC on March 24 — RSAC opens its doors at the same moment. Fear IS the product.--- DeepSeek (671B MoE)
Luc Chevry - Independent researcher
LIA - Information Services (SASU), Corbeil-Essonnes
LIA-Code : A Non-Commutative Algebra of Reasoning
IP: INPI eSoleau | DOI : zenodo.org/records/19117908
209 researchers have visited this page