{
  "_meta": {
    "name": "FrankX LLM Provider Hub",
    "description": "Curated decision layer over the frontier LLM landscape — models, benchmarks, pricing, agentic platforms, and verdicts. Built for humans and agents.",
    "url": "https://frankx.ai/llm-hub",
    "updated": "2026-06-19",
    "pricing_note": "Pricing is live via OpenRouter where available (field live=true), otherwise from the curated registry. Always verify against the provider before relying on it.",
    "sources": [
      "https://openrouter.ai/models",
      "https://artificialanalysis.ai/leaderboards/models",
      "https://lmarena.ai/",
      "https://arcprize.org/",
      "https://blog.kilo.ai/p/you-dont-have-to-use-fable-and-mythos",
      "https://github.com/frankxai/Starlight-Intelligence-System/tree/main/tools/arena/runs"
    ]
  },
  "capabilities": [
    {
      "id": "reasoning",
      "label": "Reasoning & Analysis",
      "description": "Complex problem-solving, math, abstract reasoning, long-horizon planning"
    },
    {
      "id": "multimodal",
      "label": "Multimodal Understanding",
      "description": "Vision, document, chart, and cross-modal reasoning across text/image/audio"
    },
    {
      "id": "video-gen",
      "label": "Video Generation",
      "description": "Generative video models, text-to-video, image-to-video, editing"
    },
    {
      "id": "coding",
      "label": "Coding & Engineering",
      "description": "Agentic coding, terminal use, debugging, multi-file refactors"
    },
    {
      "id": "agentic-infra",
      "label": "Agentic Infrastructure",
      "description": "Tool use, function calling, agent SDKs, computer use, long-horizon execution"
    },
    {
      "id": "voice",
      "label": "Voice & Audio",
      "description": "Native speech in/out, real-time conversation, audio understanding"
    },
    {
      "id": "image-gen",
      "label": "Image Generation",
      "description": "Text-to-image, editing, in-painting, brand-consistent generation"
    }
  ],
  "providers": [
    {
      "slug": "anthropic",
      "name": "Anthropic",
      "one_liner": "Claude frontier family for agentic coding, judgment-heavy review, and long-context work.",
      "flagship": "claude-fable-5",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "models": [
        "claude-fable-5",
        "claude-opus-4-8",
        "claude-opus-4-6",
        "claude-opus-4-5",
        "claude-sonnet-4-6",
        "claude-sonnet-4-5",
        "claude-haiku-4-5"
      ],
      "url": "https://anthropic.com"
    },
    {
      "slug": "openai",
      "name": "OpenAI",
      "one_liner": "General frontier stack for coding, voice, computer-use, and open-weight local lanes.",
      "flagship": "gpt-5-5",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal",
        "voice"
      ],
      "models": [
        "gpt-5-5",
        "gpt-5-2-pro",
        "gpt-oss"
      ],
      "url": "https://openai.com"
    },
    {
      "slug": "google",
      "name": "Google DeepMind",
      "one_liner": "Gemini and Gemma stack for agentic coding, multimodal reasoning, Antigravity, and open-weight local work.",
      "flagship": "gemini-3-5-flash",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal",
        "video-gen",
        "image-gen"
      ],
      "models": [
        "gemini-3-5-flash",
        "gemini-3-5-pro",
        "gemini-3-pro",
        "gemini-omni",
        "gemma-4"
      ],
      "url": "https://deepmind.google"
    },
    {
      "slug": "xai",
      "name": "xAI",
      "one_liner": "Grok stack for fast frontier inference, real-time context, and cost-sensitive routing.",
      "flagship": "grok-4-3",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal",
        "image-gen",
        "video-gen",
        "voice"
      ],
      "models": [
        "grok-4-3",
        "grok-4-1"
      ],
      "url": "https://x.ai"
    },
    {
      "slug": "moonshot",
      "name": "Moonshot AI",
      "one_liner": "Kimi open-weight coding lane for cost-sensitive agentic workflows.",
      "flagship": "kimi-k2-7-code",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra"
      ],
      "models": [
        "kimi-k2-6",
        "kimi-k2-7-code"
      ],
      "url": "https://www.moonshot.ai/"
    },
    {
      "slug": "minimax",
      "name": "MiniMax",
      "one_liner": "Cost-efficient coding and multimodal lane highlighted by Kilo for model-freedom routing.",
      "flagship": "minimax-m3",
      "capability_focus": [
        "coding",
        "reasoning",
        "agentic-infra",
        "multimodal",
        "video-gen"
      ],
      "models": [
        "minimax-m3"
      ],
      "url": "https://www.minimax.io/"
    },
    {
      "slug": "nvidia",
      "name": "NVIDIA",
      "one_liner": "Open-weight sovereignty lane for teams that want frontier-adjacent quality on controlled hardware.",
      "flagship": "nemotron-3-ultra",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra"
      ],
      "models": [
        "nemotron-3-ultra"
      ],
      "url": "https://www.nvidia.com/"
    },
    {
      "slug": "deepseek",
      "name": "DeepSeek",
      "one_liner": "Open-weight coding and reasoning models for low-cost and sovereign deployments.",
      "flagship": "deepseek-v4",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra"
      ],
      "models": [
        "deepseek-v4",
        "deepseek-v3-2"
      ],
      "url": "https://www.deepseek.com/"
    },
    {
      "slug": "alibaba",
      "name": "Alibaba Qwen",
      "one_liner": "Qwen closed flagship for long-context agentic and reasoning work.",
      "flagship": "qwen3-7-max",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "models": [
        "qwen3-7-max"
      ],
      "url": "https://qwen.ai/"
    },
    {
      "slug": "mistral",
      "name": "Mistral AI",
      "one_liner": "European frontier and open-weight lane for EU-sovereign AI systems.",
      "flagship": "mistral-large-3",
      "capability_focus": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "models": [
        "mistral-large-3"
      ],
      "url": "https://mistral.ai/"
    },
    {
      "slug": "meta",
      "name": "Meta AI",
      "one_liner": "Open multimodal Llama lane for self-hosted stacks and permissive deployment.",
      "flagship": "llama-4-maverick",
      "capability_focus": [
        "reasoning",
        "multimodal",
        "agentic-infra"
      ],
      "models": [
        "llama-4-maverick"
      ],
      "url": "https://ai.meta.com"
    },
    {
      "slug": "microsoft",
      "name": "Microsoft",
      "one_liner": "Small-model and edge AI lane for local, low-footprint reasoning and coding.",
      "flagship": "phi-4",
      "capability_focus": [
        "reasoning",
        "coding",
        "multimodal"
      ],
      "models": [
        "phi-4"
      ],
      "url": "https://www.microsoft.com/ai"
    }
  ],
  "models": [
    {
      "id": "claude-fable-5",
      "name": "Claude Fable 5",
      "provider": "Anthropic",
      "released": "2026-06-09",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 10,
      "output_per_1m_usd": 50,
      "pricing_live": false,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "tagline": "Mythos-class made generally available — the new agentic-coding ceiling, at 2× Opus pricing.",
      "best_for": [
        "Agentic pipelines feeding schemas, tools, and other agents (measured constraint precision)",
        "Long-horizon coding — SWE-Bench Verified 95% / Pro ~80% at launch (vendor-claimed)",
        "Hard reasoning under strict output contracts"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/claude-fable-5"
    },
    {
      "id": "claude-opus-4-8",
      "name": "Claude Opus 4.8",
      "provider": "Anthropic",
      "released": "2026-05-20",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 5,
      "output_per_1m_usd": 25,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "tagline": "Modest version bump, real frontier gains — tops the intelligence index at the same price as 4.7.",
      "best_for": [
        "Hard agentic coding and codebase-scale migrations",
        "Long-horizon autonomous work with a clear up-front spec",
        "Economically valuable knowledge work (leads GDPval-AA at 1890)"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/claude-opus-4-8"
    },
    {
      "id": "claude-opus-4-6",
      "name": "Claude Opus 4.6",
      "provider": "Anthropic",
      "released": "2026-02-05",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 5,
      "output_per_1m_usd": 25,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "tagline": "Previous reasoning + long-context flagship — superseded by Opus 4.8.",
      "best_for": [
        "Abstract reasoning (#1 ARC-AGI-2, 68.8%)",
        "Computer-use agents (#1 OSWorld, 72.7%)",
        "1M-context research synthesis and long agent sessions",
        "Parallel agent orchestration (Agent Teams)"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/claude-opus-4-6"
    },
    {
      "id": "claude-opus-4-5",
      "name": "Claude Opus 4.5",
      "provider": "Anthropic",
      "released": "2025-11-01",
      "status": "available",
      "context_tokens": 200000,
      "input_per_1m_usd": 5,
      "output_per_1m_usd": 25,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "tagline": "Previous Anthropic flagship — still strong at coding, superseded by 4.6.",
      "best_for": [
        "Coding (80.9% SWE-bench Verified)",
        "Workloads not yet migrated to 4.6"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/claude-opus-4-5"
    },
    {
      "id": "claude-sonnet-4-6",
      "name": "Claude Sonnet 4.6",
      "provider": "Anthropic",
      "released": "2026-02-05",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 3,
      "output_per_1m_usd": 15,
      "pricing_live": true,
      "capabilities": [
        "coding",
        "agentic-infra",
        "reasoning",
        "multimodal"
      ],
      "tagline": "The mid-tier that started eating the flagship’s lunch — most of Opus 4.6 at $3/$15.",
      "best_for": [
        "Production coding and integrations",
        "Content generation at scale",
        "1M-context work without flagship pricing"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/claude-sonnet-4-6"
    },
    {
      "id": "claude-sonnet-4-5",
      "name": "Claude Sonnet 4.5",
      "provider": "Anthropic",
      "released": "2025-09-29",
      "status": "available",
      "context_tokens": 1000000,
      "input_per_1m_usd": 3,
      "output_per_1m_usd": 15,
      "pricing_live": true,
      "capabilities": [
        "coding",
        "agentic-infra",
        "reasoning",
        "multimodal"
      ],
      "tagline": "The workhorse — production coding and content at mid-tier cost.",
      "best_for": [
        "Standard coding and integrations",
        "Content generation at scale",
        "Moderate-complexity production tasks"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/claude-sonnet-4-5"
    },
    {
      "id": "claude-haiku-4-5",
      "name": "Claude Haiku 4.5",
      "provider": "Anthropic",
      "released": "2025-10-01",
      "status": "available",
      "context_tokens": 200000,
      "input_per_1m_usd": 1,
      "output_per_1m_usd": 5,
      "pricing_live": true,
      "capabilities": [
        "coding",
        "agentic-infra"
      ],
      "tagline": "Fast and cheap — classification, routing, high-volume extraction.",
      "best_for": [
        "Routing and classification",
        "Real-time chat",
        "High-volume metadata tagging"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/claude-haiku-4-5"
    },
    {
      "id": "gpt-5-5",
      "name": "GPT-5.5",
      "provider": "OpenAI",
      "released": "2026-04-30",
      "status": "ga",
      "context_tokens": 1050000,
      "input_per_1m_usd": 5,
      "output_per_1m_usd": 30,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal",
        "voice"
      ],
      "tagline": "OpenAI’s agentic flagship: best-in-class computer-use and knowledge-work scores, at double the price.",
      "best_for": [
        "Terminal-agent and Codex-style autonomous loops",
        "Computer-use / OSWorld automation",
        "Long-context reasoning over large codebases and corpora"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gpt-5-5"
    },
    {
      "id": "gpt-5-2-pro",
      "name": "GPT-5.2 Pro",
      "provider": "OpenAI",
      "released": "2026-01-01",
      "status": "ga",
      "context_tokens": 400000,
      "input_per_1m_usd": 21,
      "output_per_1m_usd": 168,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "multimodal",
        "voice",
        "agentic-infra"
      ],
      "tagline": "Previous OpenAI flagship — superseded by GPT-5.5.",
      "best_for": [
        "Workloads not yet migrated to GPT-5.5",
        "Broad multimodal reasoning"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gpt-5-2-pro"
    },
    {
      "id": "gpt-oss",
      "name": "gpt-oss",
      "provider": "OpenAI",
      "released": "2026-05-01",
      "status": "ga",
      "context_tokens": 131072,
      "input_per_1m_usd": 0.04,
      "output_per_1m_usd": 0.18,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra"
      ],
      "tagline": "OpenAI’s open-weight family — Apache 2.0 reasoning that fits on one GPU or a laptop.",
      "best_for": [
        "Local-first, privacy-sensitive products (gpt-oss-20b offline on 16GB)",
        "Single-GPU reasoning workloads (gpt-oss-120b on one 80GB card)",
        "Cost-controlled agentic loops with no per-token meter or vendor lock-in"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gpt-oss"
    },
    {
      "id": "gemini-3-5-flash",
      "name": "Gemini 3.5 Flash",
      "provider": "Google DeepMind",
      "released": "2026-05-19",
      "status": "ga",
      "context_tokens": 1048576,
      "input_per_1m_usd": 1.5,
      "output_per_1m_usd": 9,
      "pricing_live": true,
      "capabilities": [
        "coding",
        "agentic-infra",
        "reasoning",
        "multimodal",
        "video-gen"
      ],
      "tagline": "Frontier agentic coding at sub-flagship economics — the new default agent runtime.",
      "best_for": [
        "High-volume agentic workloads where cost compounds",
        "MCP-heavy tool-use pipelines (83.6% MCP Atlas)",
        "Long-horizon coding agents that need 1M context cheaply"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gemini-3-5-flash"
    },
    {
      "id": "gemini-3-5-pro",
      "name": "Gemini 3.5 Pro",
      "provider": "Google DeepMind",
      "released": "2026-06-01",
      "status": "preview",
      "context_tokens": 2000000,
      "input_per_1m_usd": null,
      "output_per_1m_usd": null,
      "pricing_live": false,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal",
        "video-gen"
      ],
      "tagline": "Google’s top reasoning tier — announced, not yet shipped. Verdict pending GA.",
      "best_for": [
        "Hard reasoning where Flash hits its ceiling (once GA)",
        "Workloads needing 2M context across the widest modality set (targeted)",
        "Heavy video/audio multimodal reasoning"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gemini-3-5-pro"
    },
    {
      "id": "gemini-3-pro",
      "name": "Gemini 3 Pro",
      "provider": "Google DeepMind",
      "released": "2025-12-01",
      "status": "ga",
      "context_tokens": 2000000,
      "input_per_1m_usd": null,
      "output_per_1m_usd": null,
      "pricing_live": false,
      "capabilities": [
        "reasoning",
        "multimodal",
        "video-gen",
        "agentic-infra"
      ],
      "tagline": "Prior Gemini Pro tier — superseded by the Gemini 3.5 line (Flash GA, Pro in preview).",
      "best_for": [
        "Workloads not yet migrated to Gemini 3.5",
        "2M-context multimodal tasks"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gemini-3-pro"
    },
    {
      "id": "gemini-omni",
      "name": "Gemini Omni",
      "provider": "Google DeepMind",
      "released": "2026-05-19",
      "status": "preview",
      "context_tokens": 1000000,
      "input_per_1m_usd": null,
      "output_per_1m_usd": null,
      "pricing_live": false,
      "capabilities": [
        "video-gen",
        "multimodal",
        "agentic-infra"
      ],
      "tagline": "Native frontier video generation folded into the standard Gemini surface.",
      "best_for": [
        "Text/audio/image/video → dynamic video",
        "Natural-language video editing",
        "Enterprise post-production, virtual try-on"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gemini-omni"
    },
    {
      "id": "gemma-4",
      "name": "Gemma 4",
      "provider": "Google DeepMind",
      "released": "2026-05-01",
      "status": "ga",
      "context_tokens": 128000,
      "input_per_1m_usd": 0,
      "output_per_1m_usd": 0,
      "pricing_live": false,
      "capabilities": [
        "reasoning",
        "coding",
        "multimodal",
        "agentic-infra"
      ],
      "tagline": "Google’s open-weight flagship: a 31B frontier-tier model on one GPU, now Apache 2.0.",
      "best_for": [
        "Local-first / privacy-sensitive products (on-prem, healthcare, legal, finance)",
        "Cost-conscious agentic systems (26B A4B MoE + vLLM)",
        "Commercial fine-tuning under a clean license"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/gemma-4"
    },
    {
      "id": "grok-4-3",
      "name": "Grok 4.3",
      "provider": "xAI",
      "released": "2026-06-01",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 1.25,
      "output_per_1m_usd": 2.5,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "tagline": "Fourth-best frontier intelligence at roughly the cheapest frontier price, with the fastest output in its tier.",
      "best_for": [
        "High-volume cost-sensitive inference (classification, extraction, summarization)",
        "Latency-sensitive agentic tool loops",
        "Native video-input and voice-cloning workflows"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/grok-4-3"
    },
    {
      "id": "grok-4-1",
      "name": "Grok 4.1",
      "provider": "xAI",
      "released": "2025-11-01",
      "status": "ga",
      "context_tokens": 2000000,
      "input_per_1m_usd": null,
      "output_per_1m_usd": null,
      "pricing_live": false,
      "capabilities": [
        "reasoning",
        "multimodal"
      ],
      "tagline": "Previous Grok flagship — 2M context, superseded by Grok 4.3.",
      "best_for": [
        "Workloads not yet migrated to 4.3",
        "2M-context tasks (4.3 dropped to 1M)"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/grok-4-1"
    },
    {
      "id": "kimi-k2-6",
      "name": "Kimi K2.6",
      "provider": "Moonshot AI",
      "released": "2026-05-01",
      "status": "ga",
      "context_tokens": 262144,
      "input_per_1m_usd": 0.67,
      "output_per_1m_usd": 3.5,
      "pricing_live": true,
      "capabilities": [
        "coding",
        "reasoning",
        "agentic-infra"
      ],
      "tagline": "The open-weight model that ties GPT-5.5-class coding at one-eighth the price.",
      "best_for": [
        "Long-horizon agentic coding and Agent Swarm workloads",
        "Self-hosting near-frontier coding quality under a permissive license",
        "Cost-sensitive, high-volume agentic pipelines"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/kimi-k2-6"
    },
    {
      "id": "kimi-k2-7-code",
      "name": "Kimi K2.7 Code",
      "provider": "Moonshot AI",
      "released": "2026-06-12",
      "status": "ga",
      "context_tokens": 256000,
      "input_per_1m_usd": 0.75,
      "output_per_1m_usd": 3.5,
      "pricing_live": false,
      "capabilities": [
        "coding",
        "reasoning",
        "agentic-infra"
      ],
      "best_for": [],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/kimi-k2-7-code"
    },
    {
      "id": "minimax-m3",
      "name": "MiniMax M3",
      "provider": "MiniMax",
      "released": "2026-06-01",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 0.3,
      "output_per_1m_usd": 1.2,
      "pricing_live": false,
      "capabilities": [
        "coding",
        "reasoning",
        "agentic-infra",
        "multimodal",
        "video-gen"
      ],
      "best_for": [],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/minimax-m3"
    },
    {
      "id": "nemotron-3-ultra",
      "name": "Nemotron 3 Ultra",
      "provider": "NVIDIA",
      "released": "2026-06-01",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 0,
      "output_per_1m_usd": 0,
      "pricing_live": false,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra"
      ],
      "best_for": [],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/nemotron-3-ultra"
    },
    {
      "id": "deepseek-v4",
      "name": "DeepSeek V4",
      "provider": "DeepSeek",
      "released": "2026-05-01",
      "status": "ga",
      "context_tokens": 1048576,
      "input_per_1m_usd": 0.44,
      "output_per_1m_usd": 0.87,
      "pricing_live": true,
      "capabilities": [
        "coding",
        "reasoning",
        "agentic-infra"
      ],
      "tagline": "Open-weight frontier-class coding at one-sixth the price — MIT-licensed, 1M context, self-hostable.",
      "best_for": [
        "Budget coding agents at scale",
        "Open-weight self-hosting and fine-tuning",
        "Cost-anchor for routing decisions"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/deepseek-v4"
    },
    {
      "id": "deepseek-v3-2",
      "name": "DeepSeek V3.2",
      "provider": "DeepSeek",
      "released": "2025-12-01",
      "status": "ga",
      "context_tokens": 131072,
      "input_per_1m_usd": 0.23,
      "output_per_1m_usd": 0.34,
      "pricing_live": true,
      "capabilities": [
        "coding",
        "reasoning",
        "agentic-infra"
      ],
      "tagline": "Prior DeepSeek line — superseded by V4 (open-weight, MIT).",
      "best_for": [
        "Workloads not yet migrated to V4"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/deepseek-v3-2"
    },
    {
      "id": "qwen3-7-max",
      "name": "Qwen3.7-Max",
      "provider": "Alibaba Qwen",
      "released": "2026-05-01",
      "status": "ga",
      "context_tokens": 1000000,
      "input_per_1m_usd": 1.25,
      "output_per_1m_usd": 3.75,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "tagline": "Alibaba’s closed-weight agent flagship: top-5 intelligence, 1M context, and 35-hour autonomy at half the Western-frontier price.",
      "best_for": [
        "High-volume, verifiable long-horizon agentic workloads",
        "Teams on the Anthropic Messages protocol wanting a cheaper drop-in to trial",
        "Context-heavy agents that benefit from the 90% cached-input discount"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/qwen3-7-max"
    },
    {
      "id": "mistral-large-3",
      "name": "Mistral Large 3",
      "provider": "Mistral AI",
      "released": "2026-05-01",
      "status": "ga",
      "context_tokens": 262144,
      "input_per_1m_usd": 0.5,
      "output_per_1m_usd": 1.5,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "agentic-infra",
        "multimodal"
      ],
      "tagline": "Europe’s 675B open-weight frontier: Apache 2.0, runs on one node, sovereign by design.",
      "best_for": [
        "EU data-residency / GDPR-bound workloads",
        "Multilingual knowledge work and RAG",
        "Self-hosted frontier inference (FP8 on 8x H200)"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/mistral-large-3"
    },
    {
      "id": "llama-4-maverick",
      "name": "Llama 4 Maverick",
      "provider": "Meta AI",
      "released": "2025-12-01",
      "status": "ga",
      "context_tokens": 1048576,
      "input_per_1m_usd": 0.15,
      "output_per_1m_usd": 0.6,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "multimodal",
        "agentic-infra"
      ],
      "tagline": "Meta’s open flagship by default — permissively licensed and multimodal, but no longer leading the open pack.",
      "best_for": [
        "Permissively-licensed self-hosting in your own VPC",
        "Native open-weight multimodal (text+image) work",
        "Long-context reasoning via the 10M-token Scout sibling"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/llama-4-maverick"
    },
    {
      "id": "phi-4",
      "name": "Phi-4",
      "provider": "Microsoft",
      "released": "2025-12-01",
      "status": "ga",
      "context_tokens": 16384,
      "input_per_1m_usd": 0.07,
      "output_per_1m_usd": 0.14,
      "pricing_live": true,
      "capabilities": [
        "reasoning",
        "coding",
        "multimodal"
      ],
      "tagline": "MIT-licensed small models that punch above their weight — and run on your laptop for $0 per token.",
      "best_for": [
        "On-device and edge inference (3.8B mini in ~3-4GB VRAM)",
        "Privacy-sensitive / regulated workloads with fully local hosting",
        "High-volume, well-scoped STEM, extraction, and function-calling tasks"
      ],
      "mentioned_in": [
        "frontier-model-routing-without-fable-5"
      ],
      "url": "https://frankx.ai/llm-hub/phi-4"
    }
  ],
  "decision_matrix": [
    {
      "need": "Hardest reasoning + knowledge work",
      "pick": "claude-opus-4-8",
      "runner_up": "gpt-5-5",
      "why": "Tops the intelligence index — GDPval-AA 1890 and SWE-Bench Pro 69.2% lead the field."
    },
    {
      "need": "Agentic coding",
      "pick": "claude-fable-5",
      "runner_up": "gpt-5-5",
      "why": "New launch ceiling — 95% SWE-Bench Verified, ~80% SWE-Bench Pro vs GPT-5.5’s 58.6% (vendor-claimed)."
    },
    {
      "need": "Computer use + terminal autonomy",
      "pick": "gpt-5-5",
      "runner_up": "claude-opus-4-8",
      "why": "Best published computer-use scores (84.9% GDPval, 78.7% OSWorld, 98% Tau2 Telecom)."
    },
    {
      "need": "Lowest cost (closed frontier)",
      "pick": "grok-4-3",
      "runner_up": "gemini-3-5-flash",
      "why": "Fourth-best intelligence (AA 53) at the cheapest frontier price — $1.25/$2.50 per 1M."
    },
    {
      "need": "Top open weights",
      "pick": "kimi-k2-6",
      "runner_up": "deepseek-v4",
      "why": "Highest open-weights intelligence (AA Index 54); DeepSeek V4 is the close, MIT-licensed runner-up."
    },
    {
      "need": "Lowest cost (open weights)",
      "pick": "deepseek-v4",
      "runner_up": "gpt-oss",
      "why": "Frontier-class coding (80.6% SWE-bench Verified) at open-weight economics under MIT."
    },
    {
      "need": "Longest context",
      "pick": "grok-4-3",
      "runner_up": "gpt-5-5",
      "why": "2M-token native window; GPT-5.5 offers 1M at GA."
    },
    {
      "need": "Native voice + broad multimodal",
      "pick": "gpt-5-5",
      "runner_up": "gemini-3-5-pro",
      "why": "Native audio modality plus the widest general multimodal coverage."
    },
    {
      "need": "Widest modality (incl. video)",
      "pick": "gemini-3-5-pro",
      "runner_up": "gemini-3-5-flash",
      "why": "Google’s top reasoning tier across text/vision/audio/video (Pro in preview; Flash is the GA workhorse)."
    },
    {
      "need": "EU data sovereignty",
      "pick": "mistral-large-3",
      "runner_up": "deepseek-v4",
      "why": "Apache 2.0, EU-resident endpoints, self-hostable frontier on a single 8×H200 node."
    },
    {
      "need": "Self-host / own the weights",
      "pick": "deepseek-v4",
      "runner_up": "llama-4-maverick",
      "why": "Open-weight MoE frontier; Llama 4 for a permissive license + native multimodality."
    },
    {
      "need": "Run on one consumer GPU",
      "pick": "gemma-4",
      "runner_up": "gpt-oss",
      "why": "Gemma 4 31B runs in ~18GB at Q4 (LMArena 1452); gpt-oss-20b is the ~16GB reasoning alternative."
    },
    {
      "need": "Laptop / edge (smallest footprint)",
      "pick": "phi-4",
      "runner_up": "gemma-4",
      "why": "MIT-licensed 3.8B–15B STEM specialist that runs on a laptop; Gemma 4’s E2B/E4B tiers go smaller still."
    }
  ],
  "comparisons": [
    {
      "slug": "claude-fable-5-vs-claude-opus-4-8",
      "title": "Claude Fable 5 vs Claude Opus 4.8",
      "verdict": "Fable 5 takes agentic coding, constraint precision, and hard reasoning — at double the price. Opus 4.8 keeps situational judgment, code-craft quality, and the better $/token for human-read prose. Route by task shape, not by leaderboard.",
      "models": [
        "claude-fable-5",
        "claude-opus-4-8"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-fable-5-vs-claude-opus-4-8"
    },
    {
      "slug": "claude-fable-5-vs-gpt-5-5",
      "title": "Claude Fable 5 vs GPT-5.5",
      "verdict": "Fable 5 leads agentic coding by a generation-sized margin on launch numbers; GPT-5.5 keeps computer-use, terminal autonomy, and native voice. Different ceilings for different jobs.",
      "models": [
        "claude-fable-5",
        "gpt-5-5"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-fable-5-vs-gpt-5-5"
    },
    {
      "slug": "claude-fable-5-vs-gemini-3-5-pro",
      "title": "Claude Fable 5 vs Gemini 3.5 Pro",
      "verdict": "Not yet a fair fight: Fable 5 is generally available with published numbers; Gemini 3.5 Pro remains a limited Vertex preview with no model card, benchmarks, or pricing. Today, Fable 5 wins by forfeit — revisit at Gemini GA.",
      "models": [
        "claude-fable-5",
        "gemini-3-5-pro"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-fable-5-vs-gemini-3-5-pro"
    },
    {
      "slug": "claude-fable-5-vs-grok-4-3",
      "title": "Claude Fable 5 vs Grok 4.3",
      "verdict": "Different products. Fable 5 is the agentic-coding ceiling; Grok 4.3 is the cheapest credible frontier intelligence with the fastest throughput in its class. The 20× output-price gap means most stacks should run both — at different tiers.",
      "models": [
        "claude-fable-5",
        "grok-4-3"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-fable-5-vs-grok-4-3"
    },
    {
      "slug": "claude-fable-5-vs-deepseek-v4",
      "title": "Claude Fable 5 vs DeepSeek V4",
      "verdict": "Fable 5 owns the ceiling; DeepSeek V4 owns the open-weight floor — 80.6% SWE-Bench Verified under MIT at a tenth of the cost. If sovereignty or self-hosting is a requirement, DeepSeek wins by default; if peak agentic capability is, Fable 5 does.",
      "models": [
        "claude-fable-5",
        "deepseek-v4"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-fable-5-vs-deepseek-v4"
    },
    {
      "slug": "claude-fable-5-vs-kimi-k2-6",
      "title": "Claude Fable 5 vs Kimi K2.6",
      "verdict": "Kimi K2.6 is the best open-weights model on the neutral index and matches GPT-5.5 on SWE-Bench Pro — at $0.60/$2.50. Fable 5 still clears it by a full tier on agentic coding. Ceiling vs best-value-open: route accordingly.",
      "models": [
        "claude-fable-5",
        "kimi-k2-6"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-fable-5-vs-kimi-k2-6"
    },
    {
      "slug": "claude-fable-5-vs-qwen3-7-max",
      "title": "Claude Fable 5 vs Qwen3.7-Max",
      "verdict": "Qwen3.7-Max leads its peer group (Kimi, DeepSeek) on hard agentic coding and matches Fable 5 on context — at a quarter of the price. Fable 5 keeps a ~19-point SWE-Bench Pro lead. Value leader vs ceiling, both closed.",
      "models": [
        "claude-fable-5",
        "qwen3-7-max"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-fable-5-vs-qwen3-7-max"
    },
    {
      "slug": "claude-opus-4-8-vs-gpt-5-5",
      "title": "Claude Opus 4.8 vs GPT-5.5",
      "verdict": "Opus 4.8 leads aggregate intelligence and SWE-Bench Pro; GPT-5.5 wins computer-use, terminal-agent loops, and native voice. The split is real enough to run both.",
      "models": [
        "claude-opus-4-8",
        "gpt-5-5"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-opus-4-8-vs-gpt-5-5"
    },
    {
      "slug": "deepseek-v4-vs-claude-opus-4-8",
      "title": "DeepSeek V4 vs Claude Opus 4.8",
      "verdict": "Opus 4.8 is the stronger model outright; DeepSeek V4 is the open-weight value play — frontier-class coding at roughly a third of the price, and you can own the weights.",
      "models": [
        "deepseek-v4",
        "claude-opus-4-8"
      ],
      "url": "https://frankx.ai/llm-hub/compare/deepseek-v4-vs-claude-opus-4-8"
    },
    {
      "slug": "grok-4-3-vs-gpt-5-5",
      "title": "Grok 4.3 vs GPT-5.5",
      "verdict": "GPT-5.5 is clearly the stronger model; Grok 4.3 delivers a large share of the capability at roughly a fifth of the price — the budget-frontier default.",
      "models": [
        "grok-4-3",
        "gpt-5-5"
      ],
      "url": "https://frankx.ai/llm-hub/compare/grok-4-3-vs-gpt-5-5"
    },
    {
      "slug": "qwen3-7-max-vs-deepseek-v4",
      "title": "Qwen3.7-Max vs DeepSeek V4",
      "verdict": "Qwen3.7-Max has the higher raw intelligence but is closed and API-only; DeepSeek V4 is open-weight MIT, cheaper, and self-hostable. Capability vs control.",
      "models": [
        "qwen3-7-max",
        "deepseek-v4"
      ],
      "url": "https://frankx.ai/llm-hub/compare/qwen3-7-max-vs-deepseek-v4"
    },
    {
      "slug": "kimi-k2-6-vs-deepseek-v4",
      "title": "Kimi K2.6 vs DeepSeek V4",
      "verdict": "Kimi K2.6 edges the open-weights intelligence lead; DeepSeek V4 counters with MIT licensing, strong coding, and a deeper ecosystem. Both are self-hostable giants.",
      "models": [
        "kimi-k2-6",
        "deepseek-v4"
      ],
      "url": "https://frankx.ai/llm-hub/compare/kimi-k2-6-vs-deepseek-v4"
    },
    {
      "slug": "gpt-oss-vs-gemma-4",
      "title": "gpt-oss vs Gemma 4",
      "verdict": "Gemma 4 wins for multimodal work on a single consumer GPU; gpt-oss wins on reasoning-per-gigabyte and scales to 120b on one 80GB card. Both are Apache 2.0.",
      "models": [
        "gpt-oss",
        "gemma-4"
      ],
      "url": "https://frankx.ai/llm-hub/compare/gpt-oss-vs-gemma-4"
    },
    {
      "slug": "gemini-3-5-flash-vs-claude-opus-4-6",
      "title": "Gemini 3.5 Flash vs Claude Opus 4.6",
      "verdict": "Different tiers, different jobs. Flash wins cost-sensitive agentic coding (76.2% Terminal-Bench 2.1); Opus 4.6 wins high-stakes reasoning. Note: Opus 4.6 is now superseded by Opus 4.8 — see Opus 4.8 vs GPT-5.5 for the current flagship matchup.",
      "models": [
        "gemini-3-5-flash",
        "claude-opus-4-6"
      ],
      "url": "https://frankx.ai/llm-hub/compare/gemini-3-5-flash-vs-claude-opus-4-6"
    },
    {
      "slug": "claude-opus-4-6-vs-gpt-5-2-pro",
      "title": "Claude Opus 4.6 vs GPT-5.2 Pro",
      "verdict": "Opus 4.6 for reasoning and long-context depth; GPT-5.2 Pro for native voice and the broadest multimodal footprint. Note: both are superseded — Opus 4.8 and GPT-5.5 are the current flagships; see Opus 4.8 vs GPT-5.5.",
      "models": [
        "claude-opus-4-6",
        "gpt-5-2-pro"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-opus-4-6-vs-gpt-5-2-pro"
    },
    {
      "slug": "gemini-3-5-flash-vs-gpt-5-2-pro",
      "title": "Gemini 3.5 Flash vs GPT-5.2 Pro",
      "verdict": "Flash for cost-efficient agentic coding at scale; GPT-5.2 Pro for voice-native and broad multimodal apps. Note: GPT-5.2 Pro is superseded by GPT-5.5 — see Grok 4.3 vs GPT-5.5 or Opus 4.8 vs GPT-5.5 for current matchups.",
      "models": [
        "gemini-3-5-flash",
        "gpt-5-2-pro"
      ],
      "url": "https://frankx.ai/llm-hub/compare/gemini-3-5-flash-vs-gpt-5-2-pro"
    },
    {
      "slug": "claude-sonnet-4-5-vs-gemini-3-5-flash",
      "title": "Claude Sonnet 4.5 vs Gemini 3.5 Flash",
      "verdict": "Gemini 3.5 Flash edges ahead on agentic-coding benchmarks and cost; Claude Sonnet 4.5 remains a proven production workhorse. Note: Sonnet 4.6 (Feb 2026) is the current Anthropic mid-tier.",
      "models": [
        "claude-sonnet-4-5",
        "gemini-3-5-flash"
      ],
      "url": "https://frankx.ai/llm-hub/compare/claude-sonnet-4-5-vs-gemini-3-5-flash"
    }
  ]
}