[
    {
        "name": "Claude Opus 5.5",
        "category": "proprietary",
        "creator": "Anthropic",
        "description": "Anthropic's current flagship and the reigning leader on composite intelligence indices as of September 2026. Opus 5.5 introduces 'always-on' adaptive reasoning, dominating SWE-bench Pro and Terminal-Bench 4.0. Its 1M-token context window and unmatched agentic coding capability make it the top choice for enterprise autonomous systems and long-horizon software engineering workflows.",
        "contextWindow": "1,000,000 tokens",
        "contextValue": 1000000,
        "primaryStrength": "Agentic Coding & Intelligence",
        "sweBench": 72.0,
        "gpqa": 83.0,
        "humanEval": 97.5,
        "gaia": 74.0,
        "webarena": 62.0,
        "launch_date": "September 15, 2026",
        "costPerMillion": 4.0,
        "costLabel": "$4.00",
        "color": "from-blue-500 to-cyan-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "GPT-6 Astra",
        "category": "proprietary",
        "creator": "OpenAI",
        "description": "OpenAI's frontier reasoning model, released September 3 2026, and the peak performer in the GPT-6 family. GPT-6 Astra excels at complex research, multi-step agent reasoning, and frontier-level coding. Its 1.05M-token context window handles entire codebases in a single prompt. Best suited for high-stakes tasks where maximum capability outweighs token cost.",
        "contextWindow": "1,050,000 tokens",
        "contextValue": 1050000,
        "primaryStrength": "Frontier Reasoning & Agents",
        "sweBench": 68.0,
        "gpqa": 80.5,
        "humanEval": 96.8,
        "gaia": 70.0,
        "webarena": 58.0,
        "launch_date": "September 3, 2026",
        "costPerMillion": 10.0,
        "costLabel": "$10.00",
        "color": "from-emerald-500 to-teal-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "GPT-6 Sol",
        "category": "proprietary",
        "creator": "OpenAI",
        "description": "The balanced mid-tier in the GPT-6 family (released September 22, 2026), GPT-6 Sol delivers near-Astra capability at a fifth of the cost. With a 1.1M-token context window and strong performance on coding agent benchmarks, Sol is the go-to for production agentic workflows that demand high quality without the premium price of Astra.",
        "contextWindow": "1,100,000 tokens",
        "contextValue": 1100000,
        "primaryStrength": "Balanced Agents & Coding",
        "sweBench": 63.0,
        "gpqa": 76.0,
        "humanEval": 95.0,
        "gaia": 65.0,
        "webarena": 52.0,
        "launch_date": "September 22, 2026",
        "costPerMillion": 2.0,
        "costLabel": "$2.00",
        "color": "from-blue-500 to-indigo-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "Claude Fable 5.1",
        "category": "proprietary",
        "creator": "Anthropic",
        "description": "Anthropic's specialized long-horizon reasoning model, Claude Fable 5.1 is purpose-built for complex agentic workflows requiring sustained coherence across very large contexts. It leads on multi-step planning, instruction following, and ambiguous goal resolution — making it the preferred model for autonomous research agents and enterprise workflow orchestration.",
        "contextWindow": "500,000 tokens",
        "contextValue": 500000,
        "primaryStrength": "Long-Horizon Workflows",
        "sweBench": 65.0,
        "gpqa": 78.0,
        "humanEval": 95.5,
        "gaia": 66.0,
        "webarena": 54.0,
        "launch_date": "August 18, 2026",
        "costPerMillion": 6.0,
        "costLabel": "$6.00",
        "color": "from-purple-500 to-indigo-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "Gemini 3.8 Flash",
        "category": "proprietary",
        "creator": "Google",
        "description": "Released September 2, 2026, Gemini 3.8 Flash is Google's fastest multimodal flagship, purpose-built for high-throughput agentic workflows. Its 1M-token context window handles audio, video, images, and text natively. At $0.75/1M input tokens it is the best-value frontier model available, with pricing locked until January 2027. Top choice for fast, parallel agent sub-tasks.",
        "contextWindow": "1,048,576 tokens",
        "contextValue": 1048576,
        "primaryStrength": "Speed & Multimodal Agents",
        "sweBench": 60.0,
        "gpqa": 74.0,
        "humanEval": 94.0,
        "gaia": 62.0,
        "webarena": 49.0,
        "launch_date": "September 2, 2026",
        "costPerMillion": 0.75,
        "costLabel": "$0.75",
        "color": "from-indigo-500 to-purple-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "GPT-6 Luna",
        "category": "proprietary",
        "creator": "OpenAI",
        "description": "The high-volume efficiency model in the GPT-6 family (September 22, 2026), GPT-6 Luna delivers GPT-4-class quality at near-commodity pricing ($0.10/1M input). With a 1.1M-token context window it handles large inputs, making it ideal for classification, routing, extraction, and high-frequency sub-agent tasks where cost per call is the primary constraint.",
        "contextWindow": "1,100,000 tokens",
        "contextValue": 1100000,
        "primaryStrength": "High-Volume Cost Efficiency",
        "sweBench": 48.0,
        "gpqa": 63.0,
        "humanEval": 90.0,
        "gaia": 50.0,
        "webarena": 41.0,
        "launch_date": "September 22, 2026",
        "costPerMillion": 0.1,
        "costLabel": "$0.10",
        "color": "from-amber-500 to-orange-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "Llama 4 Maverick",
        "category": "open",
        "creator": "Meta",
        "description": "Meta's open-weight flagship from April 2025, Llama 4 Maverick uses a 400B Mixture-of-Experts architecture with only 17B active parameters per inference pass. It delivers frontier-level reasoning and vision accuracy at self-hosted cost. With a 1M-token context window and API pricing from $0.15/1M input via providers like Together AI, it is the top open-weight choice for production agentic deployments.",
        "contextWindow": "1,000,000 tokens",
        "contextValue": 1000000,
        "primaryStrength": "Open-Weight Flagship",
        "sweBench": 55.0,
        "gpqa": 69.0,
        "humanEval": 92.0,
        "gaia": 56.0,
        "webarena": 46.0,
        "launch_date": "April 15, 2025",
        "costPerMillion": 0.21,
        "costLabel": "$0.21",
        "color": "from-purple-500 to-pink-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "Llama 4 Scout",
        "category": "open",
        "creator": "Meta",
        "description": "Llama 4 Scout is Meta's long-context specialist in the open-weight space — offering a class-leading 10M-token context window on a single H100 GPU. With 109B total parameters but only 17B active via MoE, it is uniquely suited for agents that must ingest entire repositories, legal document sets, or massive retrieval corpora in one pass without chunking.",
        "contextWindow": "10,000,000 tokens",
        "contextValue": 10000000,
        "primaryStrength": "Ultra-Long Context (10M tokens)",
        "sweBench": 45.0,
        "gpqa": 60.0,
        "humanEval": 88.0,
        "gaia": 47.0,
        "webarena": 39.0,
        "launch_date": "May 20, 2025",
        "costPerMillion": 0.1,
        "costLabel": "$0.10",
        "color": "from-teal-500 to-emerald-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "Qwen3.8 235B",
        "category": "open",
        "creator": "Alibaba",
        "description": "The latest Qwen series flagship (August 2026), Qwen3.8 235B uses a 235B MoE architecture with 22B active parameters. It leads open-weight models on multilingual benchmarks and competitive coding tasks, consistently outperforming Llama 4 on math and multilingual reasoning. Available via cloud providers at ~$0.25/1M input, it is the top open-weight choice for multilingual and APAC-facing agentic applications.",
        "contextWindow": "262,144 tokens",
        "contextValue": 262144,
        "primaryStrength": "Multilingual & Math Reasoning",
        "sweBench": 52.0,
        "gpqa": 67.0,
        "humanEval": 93.0,
        "gaia": 53.0,
        "webarena": 44.0,
        "launch_date": "August 12, 2026",
        "costPerMillion": 0.25,
        "costLabel": "$0.25",
        "color": "from-purple-500 to-blue-500",
        "last_updated": "2026-09-30T00:00:00Z"
    },
    {
        "name": "Meta Muse Glimmer",
        "category": "open",
        "creator": "Meta",
        "description": "Meta's privacy-first open-weight model from the Muse family, designed specifically for self-hosted agentic workflows where data sovereignty is non-negotiable. Muse Glimmer delivers strong reasoning and coding performance optimized for on-premise deployment, making it the leading choice for regulated industries (healthcare, finance, government) that need powerful AI without sending data to third-party APIs.",
        "contextWindow": "256,000 tokens",
        "contextValue": 256000,
        "primaryStrength": "Private Self-Hosted Agents",
        "sweBench": 44.0,
        "gpqa": 58.0,
        "humanEval": 87.0,
        "gaia": 46.0,
        "webarena": 38.0,
        "launch_date": "April 8, 2026",
        "costPerMillion": 0.0,
        "costLabel": "Self-hosted",
        "color": "from-rose-500 to-pink-500",
        "last_updated": "2026-09-30T00:00:00Z"
    }
]