{"object":"hivemind.life_as_a_bee","campaign_id":"HM-S19-LIFE-AS-A-BEE-BENCHMARK","executed_at_utc":"2026-10-28T06:21:59Z","dashboard_url":"/life-as-a-bee","summary":{"total_scenarios_executed":15,"all_passed_200_ok":true,"total_tokens_processed":2291,"mean_warm_latency_ms":8077.9,"mean_tokens_per_sec":25.64,"fastest_bee_model":"hivemind/crowd-v1","fastest_bee_tokens_per_sec":71.87,"moe_speedup_vs_14b_dense":1.83,"hundred_percent_gpu_models":["hivemind/bee-stars","hivemind/bee-rusty-sundials","hivemind/bee-lean-prover","hivemind/gemma-core","hivemind/qwen35-core","hivemind/mistral-core","hivemind/crowd-v1","hivemind/coder","hivemind/coder-lite","hivemind/auto"],"sparse_moe_models":["hivemind/bee-ruche-moe","hivemind/moe-qwen35-bee","hivemind/moe-general","hivemind/moe-coder"]},"lifecycle_phases":[{"phase_id":"PHASE-1-POLLEN-INTENT","step_number":1,"title_fr":"1. Butinage d'Intention & Routage de Liquidité (Pollen Classifier)","title_en":"1. Pollen Intent Classification & Liquidity Routing","time_window":"0 – 35 ms (Temps réel)","key_metric":"15 Modèles Actifs • 0 HTTP 402/5xx","summary_fr":"Lorsqu'un développeur ou un IDE (VS Code Continue, OpenCode, Cline, Python SDK) envoie une requête, hivemind/auto analyse l'intention (code Python/Rust, EDO raide, astrophysique, preuve Lean 4, synthèse FR) et sélectionne l'abeille déjà chaude en VRAM offrant le meilleur équilibre de Nash HMT.","summary_en":"When a developer or IDE sends a prompt, hivemind/auto classifies domain intent and selects the bee already warm in producer VRAM with optimal HMT double-auction liquidity.","protocol_endpoints":["/v1/chat/completions","/v1/models","/v1/hivemind/market"]},{"phase_id":"PHASE-2-DAYTIME-NECTAR","step_number":2,"title_fr":"2. Récolte de Nectar Diurne & Préemption Jeu < 1,5 s","title_en":"2. Daytime Nectar Foraging & < 1.5s Game Preemption","time_window":"07:00 – 22:00 CET (Journée)","key_metric":"16,5 à 71,9 tok/s (100% GPU) • Yield 1,13 s","summary_fr":"Les GPU citoyens (RTX 2070 8 Go) et les gros producteurs (Tesla T4 16 Go) servent les modèles 1.5B–8B à 100% en VRAM (jusqu'à 71,9 tok/s) et les abeilles MoE 20B (2,4B–3,6B actifs) à 6,3–7,2 tok/s. Si un joueur lance un jeu Steam/3D, le bouclier local libère 100% de la VRAM en moins de 1,5 seconde.","summary_en":"Citizen GPUs (RTX 2070 8 GB) and big-GPU producers (Tesla T4 16 GB) serve 1.5B–8B bees 100% in VRAM (up to 71.9 tok/s) and 20B Sparse MoE bees at 6.3–7.2 tok/s. Launching any game preempts in < 1.5s.","protocol_endpoints":["/v1/hivemind/producer/heartbeat","/v1/hivemind/t4/status"]},{"phase_id":"PHASE-3-AMCP-WAGGLE-DANCE","step_number":3,"title_fr":"3. Danse Frétillante Inter-Abeilles (Bus AMCP v1.5 / Kafka)","title_en":"3. Inter-Bee Waggle Dance (AMCP v1.5 / Kafka Coordination)","time_window":"Coopération Multi-Agents","key_metric":"4 Topics Kafka • Échange HMT","summary_fr":"Quand une abeille généraliste ou de code rencontre un système différentiel raide (Robertson, Navier-Stokes) ou une orbite symplectique, elle délègue le sous-calcul via hivemind.amcp.bee_coord.v1 à hivemind/bee-rusty-sundials ou hivemind/bee-stars et échange de la capacité sur hivemind.amcp.bee_trading.v1.","summary_en":"Generalist and coding bees delegate stiff ODE/PDE or orbital sub-problems over hivemind.amcp.bee_coord.v1 to specialist bees (bee-rusty-sundials, bee-stars) and clear capacity on hivemind.amcp.bee_trading.v1.","protocol_endpoints":["/v1/hivemind/bees/coordinate","/v1/hivemind/bees/trade","/v1/hivemind/amcp/events"]},{"phase_id":"PHASE-4-NOCTURNAL-ROYAL-JELLY","step_number":4,"title_fr":"4. Gelée Royale Nocturne (Distillation BOINC QLoRA-MoE)","title_en":"4. Nocturnal Royal Jelly (BOINC Chained QLoRA-MoE Distillation)","time_window":"22:00 – 07:00 CET (Nuit Européenne)","key_metric":"-5,8% NLL Global • -15,1% Math","summary_fr":"La nuit, les gros producteurs GPU basculent sur des workunits BOINC QLoRA 4-bit NF4 chaînées (hmq_*). Les retours positifs du jour (+1 feedback) et les campagnes votées par la communauté affinent les experts MoE sous cliquet monotone strict (0 dérive de replay).","summary_en":"Overnight, idle GPUs execute chained 4-bit NF4 QLoRA-MoE BOINC workunits (hmq_*), distilling daytime thumbs-up traces and community-voted subject sprints under a strict monotone loss ratchet.","protocol_endpoints":["/v1/hivemind/campaigns/poll","/v1/hivemind/feedback","/v1/hivemind/usage-capture"]}],"decision_matrix":[{"use_case_fr":"Code quotidien Python / TS / SQL / Tests (VS Code, OpenCode, Zed)","use_case_en":"Everyday Python / TS / SQL coding & unit tests in IDEs","primary_bee":"hivemind/coder","backup_bee":"hivemind/coder-lite","measured_speed":"32.77 tok/s (2.93s)","gpu_residency":"100% GPU (T4 + RTX 2070)","why":"Toujours chaud en VRAM sur plusieurs GPU, génère du code propre sans préambule."},{"use_case_fr":"Éditions inline ultra-rapides, regex, one-liners shell (< 1,6s)","use_case_en":"Ultra-fast inline edits, regex, shell one-liners (< 1.6s)","primary_bee":"hivemind/coder-lite","backup_bee":"hivemind/crowd-v1","measured_speed":"53.03 tok/s (1.51s)","gpu_residency":"100% GPU (2.7 GB)","why":"2x plus rapide que les modèles 7B pour les petites transformations syntaxiques."},{"use_case_fr":"Rust système zéro-allocation, C++, algorithmique & réflexion hybride","use_case_en":"Zero-allocation Rust, C++, algorithms & hybrid thinking","primary_bee":"hivemind/qwen35-core","backup_bee":"hivemind/moe-qwen35-bee","measured_speed":"30.02 tok/s (3.66s)","gpu_residency":"100% GPU (7.4 GB)","why":"Vérifie les contraintes mémoire et de typage avant d'écrire la solution."},{"use_case_fr":"Rédaction en français, synthèse bilingue FR/EN & JSON structuré","use_case_en":"French technical writing, FR/EN bilingual synthesis & JSON","primary_bee":"hivemind/gemma-core","backup_bee":"hivemind/mistral-core","measured_speed":"36.98 tok/s (2.08s)","gpu_residency":"100% GPU (3.8 GB)","why":"Excellent français naturel, très faible coût HMT (0,6/1k) et latence de 2 secondes."},{"use_case_fr":"Astrophysique, orbites N-corps symplectiques, TTV & relativité","use_case_en":"Astrophysics, symplectic N-body orbits, exoplanet TTV & GR","primary_bee":"hivemind/bee-stars","backup_bee":"hivemind/moe-qwen35-bee","measured_speed":"27.89 tok/s (3.44s)","gpu_residency":"100% GPU (Épinglé :8294)","why":"0,0s de chargement à froid (épinglé en permanence en VRAM) et formules exactes."},{"use_case_fr":"Équations différentielles raides (EDO/EDP), Navier-Stokes, SUNDIALS","use_case_en":"Stiff ODE/PDE systems, Navier-Stokes fluids, Rusty-SUNDIALS","primary_bee":"hivemind/bee-rusty-sundials","backup_bee":"hivemind/moe-qwen35-bee","measured_speed":"33.25 tok/s (2.89s)","gpu_residency":"100% GPU (5.6 GB)","why":"Couplé à l'endpoint déterministe POST /v1/hivemind/bees/rusty-sundials/rescue."},{"use_case_fr":"Preuves formelles Lean 4 (Mathlib4) sans sorry","use_case_en":"Formal Lean 4 theorem proving (Mathlib4) without sorry","primary_bee":"hivemind/bee-lean-prover","backup_bee":"hivemind/qwen35-core","measured_speed":"16.53 tok/s (4.84s)","gpu_residency":"100% GPU (8.7 GB Q6_K)","why":"Basé sur Goedel-Prover-V2-8B quantifié en Q6_K haute précision."},{"use_case_fr":"Synthèse MoE 20B multi-experts, théorie des jeux & refactoring agentique","use_case_en":"20B Sparse MoE synthesis, game theory & agentic multi-file planning","primary_bee":"hivemind/bee-ruche-moe","backup_bee":"hivemind/moe-coder","measured_speed":"6.79–7.20 tok/s (~15.3s)","gpu_residency":"64% GPU / 36% CPU (14 GB MoE)","why":"Active seulement 2,4B à 3,6B paramètres/token : 1,8x plus rapide que le 14B dense !"}],"scenarios":[{"scenario_id":"SCEN-RUCHE-MOE-8X","scenario_title":"La Ruche 8-Expert Top-2 Sparse MoE Routing Synthesis","model_id":"hivemind/bee-ruche-moe","aliases":["bee-ruche-moe","hivemind/ruche-moe","la-ruche-moe-8x"],"category":"moe_bee","badge_label":"🐝 ABEILLE MoE 8-EXPERTS","architecture":"20B Sparse MoE (2.4B actifs / token • Top-2 sur 8 experts)","underlying_tag":"gpt-oss:20b","vram_footprint_gb":14.0,"gpu_residency":"64% GPU / 36% CPU (14.0 GB Sparse MoE)","cold_load_s":128.55,"warm_latency_ms":16205.4,"tokens_per_sec":6.79,"prompt_tokens":80,"completion_tokens":110,"total_tokens":190,"http_status":200,"domain":"general_chat","prompt":"In 2 sentences, explain why a Top-2 Sparse Mixture-of-Experts (2.4B active out of 20B total) achieves higher domain specialization per FLOP than a dense model.","response_excerpt":"[Reasoning] We need to explain in 2 sentences why a Top-2 Sparse MoE with 2.4B active out of 20B total achieves higher domain specialization per FLOP than a dense model. So we need to mention that MoE uses gating to activate only relevant experts, so each FLOP is used for domain-specific computation...","when_to_use_fr":"Requêtes interdisciplinaires combinant plusieurs domaines de La Ruche (ex. physique + code Python + synthèse en français). 1,73x plus rapide que le modèle dense 14B (coder-plus) sur T4 16 Go grâce à l'activation parcimonieuse de 2,4B paramètres par token.","when_to_use_en":"Cross-domain requests combining physics, code, and bilingual synthesis. 1.73x faster than dense 14B on a 16 GB T4 because only 2.4B parameters activate per token.","when_not_to_use":"À éviter pour l'autocomplétion temps-réel (< 2s) dans l'IDE : préférez hivemind/coder (2,9s) ou coder-lite (1,5s).","how_to_invoke":"model=\"hivemind/bee-ruche-moe\" (ou alias \"hivemind/ruche-moe\"). Conseil : passez reasoning_effort='low' et max_tokens=512 pour limiter la chaîne de pensée interne.","recommended_temperature":0.4,"recommended_max_tokens":512,"suggested_cost_per_1k":1.8},{"scenario_id":"SCEN-QWEN35-MOE-POLYGLOT","scenario_title":"Polyglot-Singularity MoE Cross-Domain PDE + Rust Synthesis","model_id":"hivemind/moe-qwen35-bee","aliases":["moe-qwen35-bee","hivemind/qwen35-moe-bee","hivemind/polyglot-singularity"],"category":"moe_bee","badge_label":"🐝 ABEILLE MoE POLYGLOTTE","architecture":"20B Sparse MoE (3.0B actifs / token • 6 experts QLoRA fusionnés)","underlying_tag":"gpt-oss:20b","vram_footprint_gb":14.0,"gpu_residency":"64% GPU / 36% CPU (14.0 GB Sparse MoE)","cold_load_s":128.55,"warm_latency_ms":17564.7,"tokens_per_sec":6.26,"prompt_tokens":54,"completion_tokens":110,"total_tokens":164,"http_status":200,"domain":"science_rusty_sundials","prompt":"Give the 1D Gross-Pitaevskii soliton conservation invariant (norm N and energy E) in 2 concise sentences.","response_excerpt":"[Reasoning] The user: \"Give the 1D Gross-Pitaevskii soliton conservation invariant (norm N and energy E) in 2 concise sentences.\" They want the expressions for the norm N and energy E for the 1D Gross-Pitaevskii soliton. Likely the standard 1D GPE: i∂ψ/∂t = -½ ∂^2ψ/∂x^2 + g|ψ|^2 ψ, with g negative for attractive interactions...","when_to_use_fr":"Problèmes scientifiques et algorithmiques polyglottes (physique quantique BEC / Gross-Pitaevskii, couplage Rust FFI + Python, vérification d'invariants mathématiques).","when_to_use_en":"Polyglot scientific and algorithmic tasks (quantum BEC solitons, Rust FFI + Python coupling, mathematical invariant checks).","when_not_to_use":"Pour un simple script mono-fichier Rust ou Python sans couplage multi-domaine, utilisez hivemind/qwen35-core (3,66s, 100% GPU).","how_to_invoke":"model=\"hivemind/moe-qwen35-bee\" (ou alias \"hivemind/polyglot-singularity\"). Idéal avec temperature=0.3 et max_tokens=600.","recommended_temperature":0.3,"recommended_max_tokens":600,"suggested_cost_per_1k":1.8},{"scenario_id":"SCEN-MOE-GENERAL-NASH","scenario_title":"Game-Theoretic Nash Equilibrium Market Clearing","model_id":"hivemind/moe-general","aliases":["moe-general","hivemind/gpt-oss-20b"],"category":"moe_bee","badge_label":"🐝 MoE RAISONNEMENT GÉNÉRAL","architecture":"20B Sparse MoE (3.6B actifs / token • Raisonnement profond)","underlying_tag":"gpt-oss:20b","vram_footprint_gb":14.0,"gpu_residency":"64% GPU / 36% CPU (14.0 GB Sparse MoE)","cold_load_s":128.55,"warm_latency_ms":15276.1,"tokens_per_sec":7.2,"prompt_tokens":62,"completion_tokens":110,"total_tokens":172,"http_status":200,"domain":"general_chat","prompt":"State in 2 concise sentences how proportional HMT credit clearing prevents free-riding in a non-monetary GPU compute market.","response_excerpt":"[Reasoning] We need to state in 2 concise sentences how proportional HMT credit clearing prevents free-riding in a non-monetary GPU compute market. The context: HMT credit clearing presumably refers to a system where credits are allocated proportionally to contributions, preventing free-riding...","when_to_use_fr":"Analyse stratégique, théorie des jeux, conception d'architecture système, synthèse académique et raisonnement logique étape par étape. Le plus rapide des 4 modèles MoE 20B (7,20 tok/s, 15,28s).","when_to_use_en":"Strategic analysis, game theory, system architecture design, and step-by-step logical reasoning. Fastest of the 20B MoE models at 7.20 tok/s.","when_not_to_use":"Si vous avez besoin d'une réponse instantanée en français sans chaîne de raisonnement longue, utilisez hivemind/gemma-core (2,08s) ou mistral-core (2,87s).","how_to_invoke":"model=\"hivemind/moe-general\" avec reasoning_effort=\"low\" ou \"medium\" et max_tokens=700.","recommended_temperature":0.5,"recommended_max_tokens":700,"suggested_cost_per_1k":1.6},{"scenario_id":"SCEN-MOE-CODER-AGENTIC","scenario_title":"Frontier Agentic Multi-File Refactoring Strategy","model_id":"hivemind/moe-coder","aliases":["moe-coder","hivemind/qwen3-coder-moe"],"category":"moe_bee","badge_label":"🐝 MoE CODE AGENTIQUE","architecture":"20B–30B Sparse MoE (3.3B actifs / token • Refactoring & Agents)","underlying_tag":"gpt-oss:20b","vram_footprint_gb":14.0,"gpu_residency":"64% GPU / 36% CPU (14.0 GB Sparse MoE)","cold_load_s":128.55,"warm_latency_ms":15848.3,"tokens_per_sec":6.94,"prompt_tokens":56,"completion_tokens":110,"total_tokens":166,"http_status":200,"domain":"coding_python","prompt":"List 3 concise invariants an autonomous coding agent must verify before hot-swapping a production router module.","response_excerpt":"[Reasoning] We need to list 3 concise invariants an autonomous coding agent must verify before hot-swapping a production router module. Likely invariants: 1) All active connections are gracefully drained or preserved 2) State/schema compatibility is verified 3) Rollback health check passes...","when_to_use_fr":"Planification agentique complexe (OpenCode, Cline, Aider), revue de sécurité multi-fichiers, conception d'invariants de hot-swap et refactoring d'architecture. 1,77x plus rapide que coder-plus (14B dense).","when_to_use_en":"Complex agentic planning (OpenCode, Cline, Aider), multi-file security audits, hot-swap invariant verification, and architectural refactoring.","when_not_to_use":"Pour générer rapidement une fonction isolée ou un test unitaire, hivemind/coder (32,77 tok/s, 2,93s) est 11x plus rapide.","how_to_invoke":"model=\"hivemind/moe-coder\" dans Aider/OpenCode pour l'étape Architecte, puis \"hivemind/coder\" pour l'étape Éditeur.","recommended_temperature":0.2,"recommended_max_tokens":800,"suggested_cost_per_1k":2.0},{"scenario_id":"SCEN-ASTRO-TTV","scenario_title":"Exoplanet Transit Timing Variations & Schwarzschild Metric","model_id":"hivemind/bee-stars","aliases":["bee-stars","hivemind-bee-stars","hivemind/astro-bee"],"category":"specialist_bee","badge_label":"🔭 ABEILLE ASTROPHYSIQUE (ÉPINGLÉE VRAM)","architecture":"7B Dense + Adaptateur QLoRA Symplectique & Astrophysique","underlying_tag":"hivemind-bee-stars","vram_footprint_gb":4.9,"gpu_residency":"100% GPU (llama-server :8294 épinglé en VRAM)","cold_load_s":0.0,"warm_latency_ms":3442.2,"tokens_per_sec":27.89,"prompt_tokens":67,"completion_tokens":96,"total_tokens":163,"http_status":200,"domain":"science_astronomy","prompt":"Give the exact formula for the Schwarzschild radius R_s and explain in 2 sentences how symplectic Velocity-Verlet preserves energy in exoplanet N-body TTV simulations.","response_excerpt":"The Schwarzschild radius \\( R_s \\) of a black hole is given by the formula: \\[ R_s = \\frac{2GM}{c^2} \\] where \\( G \\) is the gravitational constant, \\( M \\) is the mass of the black hole, and \\( c \\) is the speed of light. In exoplanet N-body TTV (Transit Timing Variations) simulations, symplectic Velocity-Verlet algorithm preserves energy...","when_to_use_fr":"Astrophysique, mécanique céleste, intégrateurs symplectiques (Velocity-Verlet, Yoshida), variations de temps de transit d'exoplanètes (TTV), relativité générale et co-pilote scientifique Unity/Steam. Zéro temps de chargement à froid (0,0s) car épinglée en permanence sur le port :8294 !","when_to_use_en":"Astrophysics, orbital mechanics, symplectic integrators (Velocity-Verlet, Yoshida), exoplanet TTV, and general relativity. 0.0s cold-start latency (permanently pinned in VRAM on :8294).","when_not_to_use":"Ne pas utiliser pour du développement web TypeScript/React ou de la bureautique générale.","how_to_invoke":"model=\"hivemind/bee-stars\" (ou alias \"hivemind/astro-bee\"). Répond immédiatement (0s cold load) à 27,89 tok/s.","recommended_temperature":0.2,"recommended_max_tokens":512,"suggested_cost_per_1k":1.1},{"scenario_id":"SCEN-SUNDIALS-ODE","scenario_title":"Stiff Robertson Chemical Kinetics BDF/Newton Rescue","model_id":"hivemind/bee-rusty-sundials","aliases":["bee-rusty-sundials","hivemind/rusty-sundials","rusty-sundials-bee"],"category":"specialist_bee","badge_label":"🔬 ABEILLE EDO/EDP & RUSTY-SUNDIALS","architecture":"7B Spécialiste Calcul Scientifique + Solveur BDF/RK45 Déterministe","underlying_tag":"qwen2.5-coder:7b-instruct","vram_footprint_gb":5.6,"gpu_residency":"100% GPU (5.6 GB VRAM)","cold_load_s":12.53,"warm_latency_ms":2887.2,"tokens_per_sec":33.25,"prompt_tokens":83,"completion_tokens":96,"total_tokens":179,"http_status":200,"domain":"science_rusty_sundials","prompt":"Explain in 3 concise bullet points why explicit RK45 fails on the stiff Robertson ODE system (k1=0.04, k2=1e4, k3=3e7) and why BDF with analytical Jacobian succeeds.","response_excerpt":"- **Explicit RK45 Fails on Stiff Robertson ODE System**: Explicit Runge-Kutta methods like RK45 are designed for non-stiff systems, where the solution changes rapidly over a short interval. The Robertson ODE system is stiff due to the large difference in time scales between the reactions...","when_to_use_fr":"Systèmes d'équations différentielles ordinaires ou partielles raides (EDO/EDP : cinétique chimique de Robertson, mécanique des fluides Navier-Stokes, solitons de Gross-Pitaevskii, génétique de Wright-Fisher) et génération de code Rust/Python utilisant SUNDIALS (CVODE, ARKODE, IDA, KINSOL).","when_to_use_en":"Stiff ODE/PDE systems (Robertson chemical kinetics, Navier-Stokes fluid mechanics, Gross-Pitaevskii quantum solitons, Wright-Fisher genetics) and Rust/Python SUNDIALS solver code.","when_not_to_use":"Pour du code CRUD standard ou de la traduction généraliste.","how_to_invoke":"Chat : model=\"hivemind/bee-rusty-sundials\" sur /v1/chat/completions (33,25 tok/s). Calcul déterministe certifié : POST /v1/hivemind/bees/rusty-sundials/rescue.","recommended_temperature":0.1,"recommended_max_tokens":600,"suggested_cost_per_1k":1.1},{"scenario_id":"SCEN-LEAN4-PROVER","scenario_title":"Formal Lean 4 Energy Ratchet Monotonicity Proof","model_id":"hivemind/bee-lean-prover","aliases":["bee-lean-prover","hivemind/goedel-prover-v2","hivemind/lean-prover"],"category":"specialist_bee","badge_label":"📐 ABEILLE PREUVE FORMELLE LEAN 4","architecture":"8B Goedel-Prover-V2 (Q6_K Haute Précision • Synthèse de Tactiques Lean 4)","underlying_tag":"hf.co/mradermacher/Goedel-Prover-V2-8B-GGUF:Q6_K","vram_footprint_gb":8.7,"gpu_residency":"100% GPU (8.7 GB VRAM Q6_K)","cold_load_s":42.27,"warm_latency_ms":4840.2,"tokens_per_sec":16.53,"prompt_tokens":57,"completion_tokens":80,"total_tokens":137,"http_status":200,"domain":"coding_general","prompt":"Write a complete Lean 4 theorem proving `theorem ratchet_le (e_parent e_child : Nat) (h : e_child ≤ e_parent) : Nat.min e_parent e_child ≤ e_parent := by` without using sorry.","response_excerpt":"[Reasoning] Okay, I need to prove that `Nat.min e_parent e_child ≤ e_parent` given that `e_child ≤ e_parent`. First, recall that `Nat.min a b` is the minimum of `a` and `b`, i.e., `Nat.min a b = a` if `a ≤ b`, and `Nat.min a b = b` if `b ≤ a`...","when_to_use_fr":"Démonstration formelle de théorèmes en Lean 4 (Mathlib4), élimination de `sorry`, synthèse de tactiques (`omega`, `linarith`, `simp`, `exact`) et vérification formelle d'invariants de sécurité ou d'énergie.","when_to_use_en":"Formal theorem proving in Lean 4 (Mathlib4), eliminating `sorry`, synthesizing tactics (`omega`, `linarith`, `simp`), and verifying security or energy invariants.","when_not_to_use":"Ne pas utiliser pour du code Python/JS classique : Goedel-Prover-V2 est spécialisé pour Lean 4 et génère une réflexion formelle préalable.","how_to_invoke":"model=\"hivemind/bee-lean-prover\" (ou \"hivemind/goedel-prover-v2\") avec max_tokens=512 et temperature=0.1.","recommended_temperature":0.1,"recommended_max_tokens":512,"suggested_cost_per_1k":1.3},{"scenario_id":"SCEN-GEMMA-BILINGUAL","scenario_title":"Synthèse Bilingue FR/EN & Souveraineté Numérique Européenne","model_id":"hivemind/gemma-core","aliases":["gemma-core","hivemind/gemma3","gemma3:4b"],"category":"core_foundation","badge_label":"🌍 CORE GEMMA 3 (GOOGLE • 4B)","architecture":"4B Google Gemma 3 Dense (128k Context • Multilingue FR/EN • 100% GPU)","underlying_tag":"gemma3:4b","vram_footprint_gb":3.8,"gpu_residency":"100% GPU (3.8 GB VRAM)","cold_load_s":32.1,"warm_latency_ms":2082.0,"tokens_per_sec":36.98,"prompt_tokens":42,"completion_tokens":77,"total_tokens":119,"http_status":200,"domain":"general_chat","prompt":"Réponds en français en 2 phrases : pourquoi l'IA citoyenne décentralisée protège-t-elle la souveraineté numérique européenne ?","response_excerpt":"L'IA citoyenne décentralisée protège la souveraineté numérique européenne en permettant aux citoyens de contrôler leurs propres données et de participer activement au développement de l'IA, sans dépendre de grandes entreprises étrangères. En distribuant le pouvoir de création et de décision, elle renforce l'autonomie et l'identité numérique de l'Europe.","when_to_use_fr":"Excellent modèle fondationnel léger (3,8 Go VRAM) pour la rédaction en français et en anglais, le résumé de longs documents, l'extraction JSON structurée et l'explication pédagogique rapide (2,08s à 36,98 tok/s).","when_to_use_en":"Fast, lightweight (3.8 GB VRAM) foundation model for French/English writing, document summarization, structured JSON extraction, and concise explanations (2.08s at 36.98 tok/s).","when_not_to_use":"Pour du code système Rust complexe ou des preuves Lean 4, préférez hivemind/qwen35-core ou hivemind/coder.","how_to_invoke":"model=\"hivemind/gemma-core\" (ou alias \"hivemind/gemma3\"). Très économe en HMT (0,6 HMT/1k) et 100% GPU.","recommended_temperature":0.5,"recommended_max_tokens":512,"suggested_cost_per_1k":0.6},{"scenario_id":"SCEN-QWEN35-RUST","scenario_title":"Hybrid Reasoning & Zero-Allocation Rust Ring Buffer","model_id":"hivemind/qwen35-core","aliases":["qwen35-core","hivemind/qwen3.5","hivemind/qwen3-8b","qwen3:8b"],"category":"core_foundation","badge_label":"🧠 CORE QWEN 3.5 (HYBRID THINKING • 8B)","architecture":"8B Qwen 3/3.5 Dense avec Raisonnement Hybride (/think & /no_think)","underlying_tag":"qwen3:8b","vram_footprint_gb":7.4,"gpu_residency":"100% GPU (7.4 GB VRAM)","cold_load_s":29.95,"warm_latency_ms":3663.9,"tokens_per_sec":30.02,"prompt_tokens":64,"completion_tokens":110,"total_tokens":174,"http_status":200,"domain":"coding_rust","prompt":"/no_think Write a minimal Rust struct `RingAvg<const N: usize>` with `push(&mut self, x: f64) -> f64` using a fixed array `[f64; N]` and zero heap allocations. Code only.","response_excerpt":"[Reasoning] Okay, I need to create a minimal Rust struct called RingAvg with a fixed-size array of N elements. The struct should have a method push that takes an f64 and returns an f64. The goal is to compute the average of the last N elements, right? And no heap allocations, so everything must be on the stack...","when_to_use_fr":"Programmation système exigeante (Rust sans allocation tas, C++, Python typé), algorithmique avec vérification d'étapes, et appels d'outils (tool-calling). Tourne à 100% en VRAM sur le T4 (30,02 tok/s, 3,66s).","when_to_use_en":"Demanding systems programming (zero-allocation Rust, C++, typed Python), algorithmic problem solving with hybrid thinking, and structured tool calling. Runs 100% in GPU VRAM at 30.02 tok/s.","when_not_to_use":"Si vous voulez uniquement le bloc de code brut sans tokens de réflexion `[Reasoning]`, utilisez hivemind/coder (2,93s).","how_to_invoke":"model=\"hivemind/qwen35-core\" (ou alias \"hivemind/qwen3.5\"). Prévoyez max_tokens >= 256 car le modèle effectue une vérification interne avant d'émettre le code final.","recommended_temperature":0.3,"recommended_max_tokens":512,"suggested_cost_per_1k":1.1},{"scenario_id":"SCEN-MISTRAL-FR-DOC","scenario_title":"Architecture Zero-Trust & Charte de Réciprocité (FR)","model_id":"hivemind/mistral-core","aliases":["mistral-core","hivemind/mistral","mistral:7b-instruct"],"category":"core_foundation","badge_label":"🇫🇷 CORE MISTRAL 7B (BILINGUE FR/EN)","architecture":"7B Mistral v0.3 Instruct Dense (Français / Anglais & Synthèse)","underlying_tag":"mistral:7b-instruct","vram_footprint_gb":6.7,"gpu_residency":"100% GPU (6.7 GB VRAM)","cold_load_s":30.29,"warm_latency_ms":2869.4,"tokens_per_sec":33.46,"prompt_tokens":56,"completion_tokens":96,"total_tokens":152,"http_status":200,"domain":"general_chat","prompt":"Explique en 3 points concis en français comment le bouclier local (Producer Shield) et la préemption en moins de 1,5 seconde protègent les joueurs sur HiveMind.","response_excerpt":"1. Le bouclier local, ou Producer Shield, est une fonctionnalité de HiveMind qui protège les producteurs locaux... 2. La préemption en moins de 1,5 seconde garantit que dès qu'un jeu est lancé, la carte graphique est immédiatement rendue au joueur sans perte de FPS...","when_to_use_fr":"Rédaction technique en français, synthèse de documentation, brainstorming, traduction FR/EN et dialogue naturel rapide (2,87s à 33,46 tok/s, 100% GPU sans préambule de réflexion).","when_to_use_en":"French technical writing, documentation synthesis, brainstorming, FR/EN translation, and direct natural dialogue (2.87s at 33.46 tok/s, 100% GPU).","when_not_to_use":"Moins spécialisé en génération de code pur que hivemind/coder ou hivemind/qwen35-core.","how_to_invoke":"model=\"hivemind/mistral-core\" (ou alias \"hivemind/mistral\") avec temperature=0.5.","recommended_temperature":0.5,"recommended_max_tokens":512,"suggested_cost_per_1k":0.9},{"scenario_id":"SCEN-CROWD-CITIZEN","scenario_title":"CrowdGPT × La Ruche Citizen AI Reciprocity Summary","model_id":"hivemind/crowd-v1","aliases":["crowd-v1","vxtzq/crowd-v1","crowdai/crowd-v1"],"category":"core_foundation","badge_label":"🤝 PARTENAIRE CROWDGPT (ULTRA-RAPIDE • 1.5B)","architecture":"1B–1.5B Dense Tronc Généraliste Citoyen (1.6 GB VRAM • 100% GPU)","underlying_tag":"qwen2.5-coder:1.5b","vram_footprint_gb":1.6,"gpu_residency":"100% GPU (1.6 GB VRAM)","cold_load_s":13.93,"warm_latency_ms":1001.7,"tokens_per_sec":71.87,"prompt_tokens":44,"completion_tokens":72,"total_tokens":116,"http_status":200,"domain":"general_chat","prompt":"Explain in 2 short sentences how a citizen GPU swarm shares idle compute without money.","response_excerpt":"A citizen GPU swarm shares idle compute without money by leveraging the power of GPUs in the cloud, which can be accessed by anyone with an internet connection. The swarm can be programmed to automatically detect idle GPUs and distribute the workload to them, without requiring any payment.","when_to_use_fr":"Modèle le plus rapide de toute la ruche (**71,87 tok/s**, **1,00s** de latence !). Parfait pour les tests de connectivité, les micro-résumés instantanés, les chatbots à très faible latence et les GPU citoyens de 2 à 4 Go de VRAM.","when_to_use_en":"Fastest model on the entire swarm (71.87 tok/s, 1.00s latency!). Ideal for health probes, instant micro-summaries, ultra-low-latency chat, and 2–4 GB VRAM citizen GPUs.","when_not_to_use":"Capacité de raisonnement limitée (1.5B) : ne pas utiliser pour des preuves mathématiques ou de l'architecture complexe.","how_to_invoke":"model=\"hivemind/crowd-v1\" (ou alias \"vxtzq/crowd-v1\"). Coût minimal : 0,3 HMT/1k.","recommended_temperature":0.4,"recommended_max_tokens":256,"suggested_cost_per_1k":0.3},{"scenario_id":"SCEN-CODER-PYTHON","scenario_title":"High-Performance Python Decorator with Microsecond Telemetry","model_id":"hivemind/coder","aliases":["coder","qwen2.5-coder:7b-instruct"],"category":"dense_coding","badge_label":"💻 PILIER CODE QUOTIDIEN (7B • RTX 2070 + T4)","architecture":"7B Qwen 2.5 Coder Instruct Dense (Multi-Producteurs Actifs)","underlying_tag":"qwen2.5-coder:7b-instruct","vram_footprint_gb":5.6,"gpu_residency":"100% GPU (5.6 GB VRAM • Servi par T4 + RTX 2070)","cold_load_s":12.53,"warm_latency_ms":2929.7,"tokens_per_sec":32.77,"prompt_tokens":68,"completion_tokens":96,"total_tokens":164,"http_status":200,"domain":"coding_python","prompt":"Write a concise Python decorator `@timed_lru` using `functools.lru_cache` and `time.perf_counter_ns` that logs execution time in microseconds. Code only.","response_excerpt":"```python\nimport functools\nimport time\n\ndef timed_lru(maxsize=None, typed=False):\n    def decorator(func):\n        @functools.wraps(func)\n        def wrapper(*args, **kwargs):\n            start_time = time.perf_counter_ns()\n            result = func(*args, **kwargs)...","when_to_use_fr":"**Le choix par défaut recommandé pour 80% du développement quotidien** dans VS Code Continue, OpenCode, Zed ou PyCharm : génération de fonctions Python/TypeScript/Rust/Go/SQL, écriture de tests pytest, correction de bugs et refactoring local. Toujours chaud en VRAM (2,93s à 32,77 tok/s).","when_to_use_en":"Recommended default for 80% of daily coding in VS Code Continue, OpenCode, Zed, or PyCharm: Python/TS/Rust functions, pytest generation, bug fixes, and local refactoring (2.93s at 32.77 tok/s).","when_not_to_use":"Pour des preuves formelles Lean 4 (utilisez bee-lean-prover) ou des systèmes d'EDO raides (utilisez bee-rusty-sundials).","how_to_invoke":"model=\"hivemind/coder\" avec max_tokens=700 et temperature=0.2 dans votre configuration IDE.","recommended_temperature":0.2,"recommended_max_tokens":700,"suggested_cost_per_1k":1.0},{"scenario_id":"SCEN-CODER-LITE-INLINE","scenario_title":"Ultra-Fast IDE TypeScript Debounce Utility","model_id":"hivemind/coder-lite","aliases":["coder-lite","qwen2.5-coder:3b"],"category":"dense_coding","badge_label":"⚡ CODE ULTRA-RAPIDE (3B • 53 TOK/S)","architecture":"3B Qwen 2.5 Coder Dense (Faible Latence IDE • 2.7 GB VRAM)","underlying_tag":"qwen2.5-coder:3b","vram_footprint_gb":2.7,"gpu_residency":"100% GPU (2.7 GB VRAM)","cold_load_s":20.32,"warm_latency_ms":1508.7,"tokens_per_sec":53.03,"prompt_tokens":60,"completion_tokens":80,"total_tokens":140,"http_status":200,"domain":"coding_web_ts","prompt":"Write a compact TypeScript `debounce<T extends (...args: any[]) => void>(fn: T, ms: number)` function. Return code only.","response_excerpt":"```typescript\nfunction debounce<T extends (...args: any[]) => void>(fn: T, ms: number): T {\n  let timeoutId: NodeJS.Timeout | null = null;\n  return function debounced(...args: Parameters<T>): void {\n    if (timeoutId) {\n      clearTimeout(timeoutId);\n    }\n    timeoutId = setTimeout(() => fn(...args), ms);\n  }","when_to_use_fr":"Petites éditions rapides dans l'IDE, génération de regex, commandes shell/git, docstrings et snippets TypeScript/Python en **1,51 seconde** à **53,03 tok/s**.","when_to_use_en":"Fast inline IDE edits, regex generation, shell/git one-liners, docstrings, and compact TypeScript/Python utilities in 1.51s at 53.03 tok/s.","when_not_to_use":"Architecture multi-fichiers ou algorithmique complexe.","how_to_invoke":"model=\"hivemind/coder-lite\" avec max_tokens=400 et temperature=0.2 (0,4 HMT/1k).","recommended_temperature":0.2,"recommended_max_tokens":400,"suggested_cost_per_1k":0.4},{"scenario_id":"SCEN-R1-ARCHITECTURE","scenario_title":"Multi-File Distributed Consensus & Byzantine Quorum Verification","model_id":"hivemind/coder-plus","aliases":["coder-plus","deepseek-r1:14b"],"category":"dense_coding","badge_label":"🏗️ DEEPSEEK-R1 14B (RAISONNEMENT DENSE)","architecture":"14B DeepSeek-R1 Distill Dense (Chaîne de Pensée Profonde)","underlying_tag":"deepseek-r1:14b","vram_footprint_gb":12.0,"gpu_residency":"75% GPU / 25% CPU (12.0 GB Dense)","cold_load_s":58.81,"warm_latency_ms":27992.6,"tokens_per_sec":3.93,"prompt_tokens":34,"completion_tokens":110,"total_tokens":144,"http_status":200,"domain":"coding_general","prompt":"In 2 concise sentences, state why Multi-Krum aggregation tolerates f Byzantine workers out of n when 2f + 2 < n.","response_excerpt":"[Reasoning] Okay, so I need to figure out why Multi-Krum aggregation can tolerate f Byzantine workers out of n when 2f + 2 < n. Hmm, I remember that Byzantine fault tolerance is about handling situations where some nodes can act maliciously or fail in arbitrary ways...","when_to_use_fr":"Vérification approfondie de protocoles distribués (Multi-Krum, consensus Byzantin), débogage de conditions de course subtiles et raisonnement mathématique dense.","when_to_use_en":"Deep verification of distributed protocols (Multi-Krum, Byzantine consensus), subtle concurrency debugging, and dense mathematical reasoning.","when_not_to_use":"Sur un GPU de 16 Go partagé, coder-plus (14B dense, 3,93 tok/s) déborde de 25% sur le CPU : les abeilles MoE 20B (`bee-ruche-moe`, `moe-coder`, `moe-general`) sont **1,75x plus rapides** (6,8–7,2 tok/s) !","how_to_invoke":"model=\"hivemind/coder-plus\" pour des tâches asynchrones de fond, ou préférez \"hivemind/moe-coder\" en interactif.","recommended_temperature":0.4,"recommended_max_tokens":800,"suggested_cost_per_1k":1.8},{"scenario_id":"SCEN-AUTO-SMART-ROUTER","scenario_title":"Smart Intent Auto-Routing (Astrophysics -> bee-stars / Code -> coder)","model_id":"hivemind/auto","aliases":["auto"],"category":"smart_router","badge_label":"🧭 ROUTEUR INTELLIGENT D'INTENTION","architecture":"Méta-Routeur d'Intention + Équilibre de Nash HMT + Résidence VRAM","underlying_tag":"smart-intent-router","vram_footprint_gb":0.0,"gpu_residency":"100% GPU (Délègue aux abeilles chaudes en VRAM)","cold_load_s":0.0,"warm_latency_ms":3056.0,"tokens_per_sec":17.67,"prompt_tokens":57,"completion_tokens":54,"total_tokens":111,"http_status":200,"domain":"science_astronomy","prompt":"Calculate the orbital period ratio of a 2:1 mean-motion exoplanet resonance and state Kepler's third law in 2 sentences.","response_excerpt":"The orbital period ratio of a 2:1 mean-motion exoplanet resonance is 2. Kepler's third law states that the square of the orbital period of a planet is directly proportional to the cube of the semi-major axis of its orbit around the Sun.","when_to_use_fr":"**Quand vous ne savez pas quelle abeille choisir !** `hivemind/auto` privilégie les modèles déjà chargés en VRAM (`loaded_in_vram: true`) pour éviter tout temps de chargement disque à froid et route automatiquement vers l'abeille disponible la plus pertinente.","when_to_use_en":"When you don't want to pick a model manually! `hivemind/auto` prioritizes models already hot in producer VRAM (`loaded_in_vram: true`) to avoid cold-load latency and routes to the best available bee.","when_not_to_use":"Si vous exigez un modèle précis (ex. syntaxe Lean 4 stricte avec `hivemind/bee-lean-prover`), appelez ce modèle explicitement.","how_to_invoke":"model=\"hivemind/auto\" — aucune configuration supplémentaire requise.","recommended_temperature":0.4,"recommended_max_tokens":512,"suggested_cost_per_1k":1.0}]}