Hub de documentation technique pour l'écosystème Cloud, IA & Labs d'infocepo.com. Ciblé sysadmins, ingénieurs cloud, développeurs et apprenants. Dernière édition: 18 mai 2026.
Mise \`a jour: 30/05/2026
Top tasks en cours de développement:
| Service | Type | Endpoint |
|---|---|---|
| AI Multilingual | LLM | api.ailab.infocepo.com |
| AI Vision | Vision | api.ailab.infocepo.com |
| AI Embedding | Embeddings | api.ailab.infocepo.com |
| AI STT | Speech-to-Text | api-audio2txt.ailab.infocepo.com |
| AI TTS Omnivoice | TTS 600 langues | api-tts-omnivoice.ailab.infocepo.com |
| Realtime AI | WebRTC WS | api-realtime-ai.ailab.infocepo.com |
| ChromaDB Vector | Base vectorielle | chromadb.ailab.infocepo.com |
| DataLab | Environnement dev | datalab.ailab.infocepo.com |
Principes clés:
Pipeline Usine IA (7 étapes):
Idea → Dev → Deploy → Monitor → Alert → Infra Support → App Support
| Critère | K8s | OpenStack | AWS | Bare-metal |
|---|---|---|---|---|
| Orchestration | ✅ | ✅ | ✅ | ❌ |
| Auto-scaling | ✅ | Moderé | ✅ | ❌ |
| Coût | Moyen | Faible | ⭐️️ | Faible |
| Contrôle total | Moderé | ✅ | ❌ | ✅ |
| IA/ML natif | ✅ | ❌ | ✅ | ✅ |
Non-production (DataLab): Expérimentation libre, snapshots fréquents
Production: Mode best-effort, monitoring continu, alertes actives
Audit cloud: Script ServerDiff.sh disponible pour migration (ex: 15 jours-homme pour migration 82 VMs)
| # | Modèle | Provider | Tokens | Tendance |
|---|---|---|---|---|
| 1 | Hy3 preview | tencent | 2.66T | ▲ +210% |
| 2 | DeepSeek V4 Flash | deepseek | 2.06T | ▲ +86% |
| 3 | Claude Sonnet 4.6 | anthropic | 1.55T | ▲ +6% |
| 4 | Claude Opus 4.7 | anthropic | 1.54T | ▲ +24% |
| 5 | Gemini 3 Flash Preview | 1.15T | ▲ +7% | |
| 6 | Kimi K2.6 | moonshotai | 1.05T | ▼ -35% |
| 7 | DeepSeek V3.2 | deepseek | 1.03T | ▲ +18% |
| 8 | Owl Alpha | openrouter | 895B | ▲ +121% |
| 9 | DeepSeek V4 Pro | deepseek | 893B | ▲ +9% |
| 10 | MiniMax M2.7 | minimax | 750B | ▲ +1% |
| # | Provider | Volume | Part |
|---|---|---|---|
| 1 | DeepSeek | 689B | 19.3% |
| 2 | 522B | 14.6% | |
| 3 | Anthropic | 489B | 13.7% |
| 4 | Tencent | 433B | 12.1% |
| 5 | OpenAI | 368B | 10.3% |
| 6 | OpenRouter | 169B | 4.7% |
| 7 | Moonshot AI | 146B | 4.1% |
| 8 | Qwen | 145B | 4.1% |
| 9 | Z-AI (GLM) | 116B | 3.2% |
| # | Modèle | Score |
|---|---|---|
| 1 | GPT-5.5 (xhigh) | 60.2 |
| 2 | Claude Opus 4.7 (Adaptive) | 57.3 |
| 3 | MiMo-V2.5-Pro | 53.8 |
| 4 | Grok 4.3 | 53.2 |
| 5 | GPT-5 Codex (high) | 44.6 |
| 6 | Qwen3.6 35B A3B | 43.5 |
| 7 | MiniMax-M2.1 | 39.4 |
| 8 | Mistral Medium 3.5 | 39.2 |
| # | Modèle | Débit | Prix/M tokens |
|---|---|---|---|
| 1 | gpt-oss-120b (Cerebras) | 658 tok/s | $0.35 |
| 2 | gpt-oss-safeguard-20b (Groq) | 614 tok/s | $0.07 |
| 3 | Qwen3 32B (Groq) | 440 tok/s | $0.29 |
| 4 | gpt-oss-20b (Groq) | 415 tok/s | $0.07 |
| 5 | Mercury 2 (Inception) | 331 tok/s | $0.25 |
| 6 | Llama 3.1 8B (Cerebras) | 270 tok/s | $0.10 |
Plus populaires:
Gemma4 (~9.2M pulls) > Qwen3.5 (~11.4M) > Nemotron3 (582K) > Kimi K2.6 (250K)
Range paramètres: 0.8B .. 744B total / 40B actif (GLM-5)
| # | Modèle | Créateur | Tâche | Likes |
|---|---|---|---|---|
| 1 | openbmb/MiniCPM-V-4.6 | openbmb | Image-Text | 769 |
| 2 | Sulphur-2-base | SulphurAI | Text-to-Video | 1,120 |
| 3 | supertonic-3 | Supertone | Text-to-Speech | 415 |
| 4 | Qwen3.6-27B-MTP-GGUF | unsloth | Image-Text | 284 |
| 5 | Qwen3.6-35B-A3B-MTP-GGUF | unsloth | Image-Text | 247 |
| 6 | Anima | circlestone | — | 1,400 |
| 7 | Dramabox | ResembleAI | Text-to-Speech | 158 |
| 8 | DeepSeek-V4-Pro | deepseek-ai | Text Gen | 4,040 |
| 9 | HiDream-O1-Image | HiDream-ai | Img-to-Img | 390 |
| 10 | ZAYA1-8B | Zyphra | — | 530 |
| 11 | supergemma4-26b-gguf | Jiunsong | Text Gen | 626 |
| 12 | DeepSeek-V4-Flash | deepseek-ai | Text Gen | 1,150 |
| 13 | Qwen3.6-35B-A3B | Qwen | Image-Text | 1,820 |
| 14 | Qwen3.6-27B | Qwen | Image-Text | 1,330 |
| 15 | gemma-4-31B-it | Image-Text | 2,680 |
| # | Modèle | Téléch. | Likes |
|---|---|---|---|
| 1 | all-MiniLM-L6-v2 | 258.98M | 4,800 |
| 2 | Qwen3-VL-2B-Instruct | 137.94M | 406 |
| 3 | bert-base-uncased | 67.67M | 2,655 |
| 4 | ms-marco-MiniLM-L6-v2 | 55.11M | 240 |
| 5 | electra-base-discriminator | 54.72M | 107 |
| 6 | paraphrase-multilingual-MiniLM | 48.10M | 1,227 |
| 7 | bge-small-en-v1.5 | 45.72M | 463 |
| 8 | all-mpnet-base-v2 | 35.49M | 1,292 |
| 9 | clip-vit-large-patch14 | 32.83M | 2,013 |
| 10 | bge-m3 | 26.60M | 3,011 |
| # | Répo | Lang | ⭐ Stars |
|---|---|---|---|
| 1 | NousResearch/hermes-agent | Python | 156k |
| 2 | anomalyco/opencode | TypeScript | 162k |
| 3 | langflow-ai/langflow | Python | 148k |
| 4 | langchain-ai/langchain | Python | 137k |
| 5 | firecrawl/firecrawl | TypeScript | 121k |
| 6 | google-gemini/gemini-cli | TypeScript | 104k |
| 7 | openai/codex | Rust | 83.5k |
| 8 | warpdotdev/warp | Rust | 59k |
| 9 | earendil-works/pi | TypeScript | 51.2k |
| 10 | ansible/ansible | Python | 68.6k |
| # | Répo | Lang | ⭐ Stars |
|---|---|---|---|
| 1 | openclaw/openclaw | TypeScript | 373k |
| 2 | n8n-io/n8n | TypeScript | 189k |
| 3 | ollama/ollama | Go | 172k |
| 4 | NousResearch/hermes-agent | Python | 156k |
| 5 | langgenius/dify | Python | 142k |
| 6 | x1xhlol/system-prompts-ai-tools | — | 138k |
| 7 | open-webui/open-webui | Python | 138k |
| 8 | firecrawl/firecrawl | TypeScript | 121k |
| 9 | Snailclimb/JavaGuide | Java | 156k |
| 10 | langflow-ai/langflow | Python | 148k |
| # | Répo | Description | Lang | ⭐ | Forks |
|---|---|---|---|---|---|
| 1 | CloakHQ/CloakBrowser | Stealth Chromium qui passe tous les tests bot detection | Python | 15,093 | 1,176 |
| 2 | rohitg00/agentmemory | Mémoire persistante pour agents IA de codage | TypeScript | 12,696 | 1,073 |
| 3 | oven-sh/bun | Runtime JS ultra-rapide, bundler, test runner | Rust | 91,928 | 4,595 |
| 4 | Imbad0202/academic-research-skills | Compétences recherche académique Claude Code | Python | 11,520 | 1,181 |
| 5 | yikart/AiToEarn | Utilisons l'IA pour gagner! | TypeScript | 15,174 | 2,490 |
| 6 | anthropics/financial-services | Outils services financiers Anthropic | Python | 25,298 | 3,498 |
| 7 | mattpocock/skills | Skills for Real Engineers | Shell | 91,684 | 8,037 |
| 8 | ruvnet/RuView | WiFi signals → spatial intelligence & vital signs | Rust | 59,834 | 7,806 |
| 9 | millionco/react-doctor | Détecte le mauvais code React généré par agent | TypeScript | 10,159 | 324 |
| 10 | colbymchenry/codegraph | Graph de connaissances code local pour agents IA | TypeScript | 4,764 | 336 |
| 11 | apernet/hysteria | Proxy rapide et résistant \`a la censure | Go | 21,233 | 2,180 |
| 12 | facebook/pyrefly | Vérificateur de type Python rapide | Rust | 6,199 | 367 |
| 13 | bytedance/UI-TARS-desktop | Pile IA multimodale open-source | TypeScript | 34,615 | 3,470 |
Classement basé sur 305,461 votes utilisateurs • 79 modèles évalués • Dernière mise \`a jour: 14 mai 2026
| # | Modèle | Score | +/- | Provider |
|---|---|---|---|---|
| 1 | claude-opus-4-7-thinking | 1567 | +11/-11 | Anthropic |
| 2 | claude-opus-4-7 | 1559 | +11/-11 | Anthropic |
| 3 | claude-opus-4-6-thinking | 1546 | +8/-8 | Anthropic |
| 4 | claude-opus-4-6 | 1541 | +8/-8 | Anthropic |
| # | Modèle | Score |
|---|---|---|
| 1 | GPT-5.5 (xhigh) | 60.24 |
| 2 | Claude Opus 4.7 (max) | 57.28 |
| 3 | Gemini 3.1 Pro Preview | 57.18 |
| 4 | GPT-5.4 (xhigh) | 56.80 |
| 5 | Kimi K2.6 | 53.90 |
| 6 | MiMo-V2.5-Pro | 53.83 |
| 7 | Grok 4.3 (high) | 53.20 |
| 8 | Muse Spark | 52.15 |
| 9 | Qwen3.6 Max Preview | 51.81 |
| 10 | Claude Sonnet 4.6 (max) | 51.72 |
| # | Modèle | Débit | Prix/M |
|---|---|---|---|
| 1 | gpt-oss-120B (high) | 249 tok/s | $0.26 |
| 2 | gpt-oss-20B (high) | 242 tok/s | $0.09 |
| 3 | NVIDIA Nemotron 3 Super | 225 tok/s | $0.41 |
| 4 | GPT-5.4 mini (xhigh) | 168 tok/s | $1.69 |
| 5 | Gemini 3 Flash | 165 tok/s | $1.13 |
| 6 | Mistral Medium 3.5 | 155 tok/s | $3.00 |
| 7 | Nova 2.0 Pro Preview | 135 tok/s | $3.44 |
| 8 | Gemini 3.1 Pro Preview | 130 tok/s | $4.50 |
| 9 | Grok 4.3 (high) | 102 tok/s | $1.56 |
| 10 | Claude 4.5 Haiku | 101 tok/s | $2.19 |
| # | Modèle | Provider | Prix/M | Débit |
|---|---|---|---|---|
| 1 | DeepSeek V4 Flash | DeepSeek | $0.18 | 97 tok/s |
| 2 | MiniMax-M2.7 | MiniMax | $0.52 | 47 tok/s |
| 3 | NVIDIA Nemotron 3 Super | NVIDIA | $0.41 | 225 tok/s |
| 4 | DeepSeek V3.2 | DeepSeek | $0.34 | — |
| 5 | Qwen3.5 397B A17B | Alibaba | $1.35 | 53 tok/s |
| 6 | Grok 4.3 | xAI | $1.56 | 102 tok/s |
| 7 | GPT-5.4 mini | OpenAI | $1.69 | 168 tok/s |
| 8 | Kimi K2.6 | Moonshot AI | $1.71 | 98 tok/s |
| # | Modèle | Contexte | Provider |
|---|---|---|---|
| 1 | DeepSeek V4 Flash | 1,000,000 | DeepSeek |
| 2 | DeepSeek V4 Pro | 1,000,000 | DeepSeek |
| 3 | Claude Opus 4.7 | 1,000,000 | Anthropic |
| 4 | Claude Sonnet 4.6 | 1,000,000 | Anthropic |
| 5 | Gemini 3 Flash | 1,000,000 | |
| 6 | Gemini 3.1 Pro | 1,000,000 | |
| 7 | NVIDIA Nemotron 3 Super | 1,000,000 | NVIDIA |
| 8 | GPT-5.4 | 1,050,000 | OpenAI |
1. MoE & Efficiency: Les architectures Mixture-of-Experts dominent (DeepSeek V4, Qwen3.6, Nemotron 3 Super). Activation de sous-ensembles de paramètres pour un coût/réponse optimal.
2. Agentic Coding: Hermes Agent, Kilo Code, OpenCode CLI, Gemini CLI — les agents autonomes de codage explosent sur GitHub.
3. Multimodal Standard: Vision + audio + texte intégrés dans les modèles flagship (Gemma4, Qwen3.6, Kimi K2.6).
4. Video Generation: Sulphur-2, LTX-2.3 — la génération vidéo open-source accélère.
5. Speed Wars: Groq et Cerebras dominent la vitesse (>600 tok/s), NVIDIA Nemotron 3 Super offre le meilleur rapport vitesse/coût.
6. China Rise: Tencent (Hy3), DeepSeek, Moonshot AI (Kimi), Z-AI (GLM), Alibaba (Qwen) — forte présence chinoise dans le top mondial.
7. Context Windows: 1M tokens devient la norme pour les modèles frontier (DeepSeek, Anthropic, Google).