[
    {
        "provider": "anthropic",
        "naming": "claude-{tier}-{major}-{minor} — tiers: haiku < sonnet < opus < fable/mythos (premium); claude-{major}-{tier}-{date} (3.x legacy)",
        "rateSource": "https://platform.claude.com/docs/en/about-claude/pricing",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "claude",
            "supportsEffort": true,
            "inputTokenSoftLimit": 150000
        },
        "profiles": [
            {
                "prefix": "claude",
                "tier": "standard",
                "cost": "medium",
                "input": 3,
                "output": 15,
                "cachedRead": 0.3,
                "cachedNonRead": 3.75,
                "bestFor": "general Claude tasks (catch-all, Sonnet rates)"
            },
            {
                "prefix": "claude-opus",
                "tier": "reasoning",
                "cost": "high",
                "input": 5,
                "output": 25,
                "cachedRead": 0.5,
                "cachedNonRead": 6.25,
                "bestFor": "high-ambiguity tasks where error cost is high (flagship: claude-opus-4-8, released 2026-05-28, 1M context)"
            },
            {
                "prefix": "claude-sonnet",
                "bestFor": "professional coding and tool use (Sonnet 4.6 rates)"
            },
            {
                "prefix": "claude-sonnet-5",
                "input": 3,
                "output": 15,
                "cachedRead": 0.3,
                "cachedNonRead": 3.75,
                "bestFor": "current Sonnet flagship (Sonnet 5, GA 2026-06-30); default on Free/Pro; standard $3/$15 — introductory $2/$10 applied through 2026-08-31"
            },
            {
                "prefix": "claude-fable",
                "tier": "reasoning",
                "cost": "high",
                "input": 10,
                "output": 50,
                "cachedRead": 1,
                "cachedNonRead": 12.5,
                "bestFor": "premium Claude tier above Opus for highest-stakes creative + reasoning work (Fable 5; $10/$50)"
            },
            {
                "prefix": "claude-mythos",
                "tier": "reasoning",
                "cost": "high",
                "input": 10,
                "output": 50,
                "cachedRead": 1,
                "cachedNonRead": 12.5,
                "bestFor": "premium Claude reasoning tier (Mythos 5; same $10/$50 as Fable 5)"
            },
            {
                "prefix": "claude-haiku",
                "cost": "low",
                "input": 1,
                "output": 5,
                "cachedRead": 0.1,
                "cachedNonRead": 1.25,
                "bestFor": "fast low-cost coding and batch work"
            },
            {
                "prefix": "claude-opus-4-1",
                "input": 15,
                "output": 75,
                "cachedRead": 1.5,
                "cachedNonRead": 18.75,
                "bestFor": "active; tentative retirement not sooner than 2026-08-05"
            },
            {
                "prefix": "claude-3",
                "supportsEffort": false,
                "bestFor": "legacy 3.x (retired)"
            },
            {
                "prefix": "claude-3-opus",
                "tier": "reasoning",
                "cost": "high",
                "input": 15,
                "output": 75,
                "cachedNonRead": 18.75,
                "bestFor": "RETIRED 2026-01-05 — calls will fail; rates retained for historical cost accounting"
            },
            {
                "prefix": "claude-3-haiku",
                "cost": "low",
                "input": 0.25,
                "output": 1.25,
                "cachedNonRead": 0.3125,
                "bestFor": "RETIRED 2026-04-20 — calls will fail; rates retained for historical cost accounting"
            }
        ]
    },
    {
        "provider": "openai",
        "naming": "gpt-{major}.{minor}[-{size}] — 5.6 sizes: sol (flagship), terra (mid), luna (budget); 5.4/5.5 sizes: pro, mini, nano (no suffix = base/flagship)",
        "rateSource": "https://developers.openai.com/api/docs/pricing",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "openai",
            "supportsEffort": true,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "gpt",
                "tier": "standard",
                "cost": "medium",
                "input": 2.5,
                "output": 15,
                "cachedRead": 0.25,
                "bestFor": "general GPT family tasks (catch-all)"
            },
            {
                "prefix": "gpt-5",
                "tier": "reasoning",
                "bestFor": "reasoning-tier coding and agentic workflows",
                "inputTokenSoftLimit": 1000000
            },
            {
                "prefix": "gpt-5.6",
                "input": 5,
                "output": 30,
                "cachedRead": 0.5,
                "bestFor": "current GPT flagship (GPT-5.6 Sol, GA 2026-07-09); base id resolves to the Sol tier; cache writes bill at 1.25x uncached input"
            },
            {
                "prefix": "gpt-5.6-sol",
                "input": 5,
                "output": 30,
                "cachedRead": 0.5,
                "bestFor": "GPT-5.6 Sol flagship tier"
            },
            {
                "prefix": "gpt-5.6-terra",
                "input": 2.5,
                "output": 15,
                "cachedRead": 0.25,
                "bestFor": "GPT-5.6 Terra mid tier"
            },
            {
                "prefix": "gpt-5.6-luna",
                "tier": "standard",
                "cost": "low",
                "input": 1,
                "output": 6,
                "cachedRead": 0.1,
                "bestFor": "GPT-5.6 Luna budget tier"
            },
            {
                "prefix": "gpt-5.5",
                "input": 5,
                "output": 30,
                "cachedRead": 0.5,
                "bestFor": "prior GPT-5 flagship (superseded by GPT-5.6 2026-07-09; still served)"
            },
            {
                "prefix": "gpt-5.5-pro",
                "cost": "high",
                "input": 30,
                "output": 180,
                "bestFor": "highest-quality reasoning and complex research"
            },
            {
                "prefix": "gpt-5.4",
                "cachedRead": 0.25,
                "bestFor": "prior GPT-5 flagship"
            },
            {
                "prefix": "gpt-5.4-pro",
                "cost": "high",
                "input": 30,
                "output": 180,
                "bestFor": "previous-gen high-end reasoning"
            },
            {
                "prefix": "gpt-5.4-mini",
                "tier": "standard",
                "cost": "low",
                "input": 0.75,
                "output": 4.5,
                "cachedRead": 0.075,
                "bestFor": "fast cost-effective tasks and extraction"
            },
            {
                "prefix": "gpt-5.4-nano",
                "tier": "standard",
                "cost": "low",
                "input": 0.2,
                "output": 1.25,
                "cachedRead": 0.02,
                "bestFor": "cheapest GPT for high-throughput batch work"
            },
            {
                "prefix": "gpt-5.3",
                "input": 1.75,
                "output": 14,
                "cachedRead": 0.175,
                "bestFor": "reasoning and coding at lower cost than 5.4 (gpt-5.3-codex; the gpt-5.x-codex / -chat-latest snapshots retire 2026-07/08)"
            },
            {
                "prefix": "gpt-4",
                "tier": "standard",
                "supportsEffort": false,
                "bestFor": "legacy GPT-4 family"
            }
        ]
    },
    {
        "provider": "google",
        "naming": "gemini-{major}[.{minor}]-{tier}-{suffix} — tiers: pro ($2/$12), flash ($0.50/$3)",
        "rateSource": "https://ai.google.dev/pricing",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "gemini",
            "supportsEffort": false,
            "inputTokenSoftLimit": 200000
        },
        "profiles": [
            {
                "prefix": "gemini",
                "tier": "standard",
                "cost": "low",
                "input": 0.5,
                "output": 3,
                "bestFor": "general Gemini tasks (catch-all, Flash rates)"
            },
            {
                "prefix": "gemini-3-flash",
                "tier": "standard",
                "cost": "low",
                "input": 0.5,
                "output": 3,
                "cachedRead": 0.05,
                "bestFor": "Gemini 3 Flash (preview; being phased out in favor of gemini-3.5-flash)"
            },
            {
                "prefix": "gemini-3.5-flash",
                "tier": "standard",
                "cost": "medium",
                "input": 1.5,
                "output": 9,
                "cachedRead": 0.15,
                "bestFor": "current Gemini Flash flagship (GA 2026-05-19); stronger than 3.x Flash at higher rates"
            },
            {
                "prefix": "gemini-3.1-flash-lite",
                "tier": "standard",
                "cost": "low",
                "input": 0.25,
                "output": 1.5,
                "cachedRead": 0.025,
                "bestFor": "cheapest Gemini tier for high-throughput tasks"
            },
            {
                "prefix": "gemini-3.1-pro",
                "tier": "reasoning",
                "cost": "medium",
                "input": 2,
                "output": 12,
                "cachedRead": 0.2,
                "notes": "tiered: $2/$12 ≤200k input tokens, $4/$18 above. Cost meter assumes ≤200k tier — long-context calls may underestimate.",
                "bestFor": "long-context reasoning with large documents (current Gemini Pro flagship, 1M context)",
                "inputTokenSoftLimit": 1000000
            },
            {
                "prefix": "gemini-3-pro",
                "tier": "reasoning",
                "cost": "medium",
                "input": 2,
                "output": 12,
                "notes": "tiered: $2/$12 ≤200k input tokens, $4/$18 above.",
                "bestFor": "DISCONTINUED 2026-03-26 — replaced by gemini-3.1-pro; rates retained for historical cost accounting"
            },
            {
                "prefix": "gemini-2.5-pro",
                "tier": "reasoning",
                "cost": "medium",
                "input": 1.25,
                "output": 10,
                "notes": "tiered: $1.25/$10 ≤200k input tokens, $2.50/$15 above.",
                "bestFor": "reasoning and coding with extended thinking"
            },
            {
                "prefix": "gemini-2.5-flash",
                "tier": "standard",
                "cost": "low",
                "input": 0.3,
                "output": 2.5,
                "bestFor": "Gemini 2.5 Flash (legacy generation)"
            }
        ]
    },
    {
        "provider": "xai",
        "naming": "grok-{major}[.{minor}][-{variant}] — Grok 4.3 (slug grok-4.3) is current flagship; grok-4.20 is the 2M-context variant; grok-4 family (incl. heavy) deprecated 2026-05-15, retires 2026-08-15",
        "rateSource": "https://docs.x.ai/developers/models",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "grok",
            "supportsEffort": false,
            "inputTokenSoftLimit": 200000
        },
        "profiles": [
            {
                "prefix": "grok",
                "tier": "reasoning",
                "cost": "low",
                "input": 1.25,
                "output": 2.5,
                "cachedRead": 0.2,
                "bestFor": "general Grok tasks (catch-all, Grok 4.3 rates)"
            },
            {
                "prefix": "grok-4-3",
                "tier": "reasoning",
                "cost": "low",
                "input": 1.25,
                "output": 2.5,
                "cachedRead": 0.2,
                "bestFor": "current Grok flagship (dash-form alias of grok-4.3)",
                "inputTokenSoftLimit": 1000000
            },
            {
                "prefix": "grok-4.3",
                "tier": "reasoning",
                "cost": "low",
                "input": 1.25,
                "output": 2.5,
                "cachedRead": 0.2,
                "bestFor": "current Grok flagship (released 2026-04-30); 1M context; retired grok-4 / grok-4-fast slugs redirect here",
                "inputTokenSoftLimit": 1000000
            },
            {
                "prefix": "grok-4.20",
                "tier": "reasoning",
                "cost": "low",
                "input": 2,
                "output": 6,
                "cachedRead": 0.2,
                "bestFor": "Grok 4.20 reasoning / non-reasoning / multi-agent variants; 2M context",
                "inputTokenSoftLimit": 2000000
            },
            {
                "prefix": "grok-4.1-fast",
                "tier": "reasoning",
                "cost": "low",
                "input": 0.2,
                "output": 0.5,
                "cachedRead": 0.05,
                "bestFor": "cheapest frontier-tier Grok (Grok 4.1 Fast); high-throughput budget option"
            },
            {
                "prefix": "grok-4-heavy",
                "tier": "reasoning",
                "cost": "high",
                "input": 15,
                "output": 75,
                "bestFor": "DEPRECATED 2026-05-15, retires 2026-08-15 — heavy mode now lives inside grok-4.3 multi-agent; rates retained for historical cost accounting",
                "inputTokenSoftLimit": 256000
            },
            {
                "prefix": "grok-build",
                "tier": "standard",
                "cost": "low",
                "input": 1,
                "output": 2,
                "bestFor": "coding-tuned Grok (grok-build-0.1, successor to grok-code-fast-1); 256k context",
                "inputTokenSoftLimit": 256000
            }
        ]
    },
    {
        "provider": "mistral",
        "naming": "mistral-{tier}-{version} — tiers: large ($0.50/$1.50), small ($0.15/$0.60)",
        "rateSource": "https://mistral.ai/pricing",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "mistral",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "mistral",
                "tier": "standard",
                "cost": "low",
                "input": 0.5,
                "output": 1.5,
                "bestFor": "general Mistral tasks (catch-all, Large 3 rates)"
            },
            {
                "prefix": "mistral-large",
                "bestFor": "cost-effective frontier-class coding (Large 3)"
            },
            {
                "prefix": "mistral-medium",
                "cost": "medium",
                "input": 1.5,
                "output": 7.5,
                "bestFor": "mid-tier Mistral (Medium 3.5; price rose from Medium 3's $0.40/$2.00)"
            },
            {
                "prefix": "mistral-small",
                "input": 0.15,
                "output": 0.6,
                "bestFor": "fast lightweight tasks and classification (Small 4)"
            },
            {
                "prefix": "codestral",
                "family": "mistral",
                "tier": "standard",
                "cost": "low",
                "input": 0.3,
                "output": 0.9,
                "bestFor": "coding-tuned Mistral"
            },
            {
                "prefix": "devstral",
                "family": "mistral",
                "tier": "standard",
                "cost": "low",
                "input": 0.4,
                "output": 2,
                "bestFor": "coding-agent Mistral line (Devstral 2 / devstral-medium rates)"
            },
            {
                "prefix": "devstral-small",
                "input": 0.1,
                "output": 0.3,
                "bestFor": "small coding-agent Mistral (Devstral Small 2)"
            },
            {
                "prefix": "ministral",
                "family": "mistral",
                "tier": "standard",
                "cost": "low",
                "input": 0.2,
                "output": 0.2,
                "bestFor": "edge-sized Mistral models (Ministral 3B/8B/14B; 14B rates)"
            }
        ]
    },
    {
        "provider": "deepseek",
        "naming": "deepseek-{variant}[-{size}] — V4 sizes: flash, pro; legacy variants are flat-priced",
        "rateSource": "https://api-docs.deepseek.com/quick_start/pricing",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "deepseek",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "deepseek",
                "tier": "standard",
                "cost": "low",
                "input": 0.14,
                "output": 0.28,
                "bestFor": "cost-effective coding and reasoning (catch-all, v4-flash rates)"
            },
            {
                "prefix": "deepseek-v4-flash",
                "input": 0.14,
                "output": 0.28,
                "cachedRead": 0.0028,
                "bestFor": "cheapest DeepSeek V4 with 1M context",
                "inputTokenSoftLimit": 1000000
            },
            {
                "prefix": "deepseek-v4-pro",
                "tier": "reasoning",
                "cost": "medium",
                "input": 0.435,
                "output": 0.87,
                "cachedRead": 0.003625,
                "bestFor": "DeepSeek V4 thinking-mode reasoning with 1M context (the 75%-off launch rate was made permanent 2026-05-22)",
                "inputTokenSoftLimit": 1000000
            }
        ]
    },
    {
        "provider": "meta",
        "naming": "llama-{major}-{variant} | meta-llama/llama-{major}-{variant} (OpenRouter) — Llama 4 (Maverick/Scout) is current; no Llama 5 as of 2026-06-01",
        "rateSource": "https://openrouter.ai/meta-llama",
        "rateLookupDate": "2026-06-01",
        "defaults": {
            "family": "llama",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "llama",
                "tier": "standard",
                "cost": "low",
                "input": 0.15,
                "output": 0.6,
                "bestFor": "general open-weight model tasks"
            },
            {
                "prefix": "llama-4",
                "bestFor": "cost-effective open-weight coding"
            },
            {
                "prefix": "meta-llama/",
                "tier": "standard",
                "cost": "low",
                "input": 0.15,
                "output": 0.6,
                "bestFor": "open-weight coding via OpenRouter"
            }
        ]
    },
    {
        "provider": "alibaba",
        "naming": "qwen{major}[.{minor}]-{tier} — no separator after 'qwen'; Qwen3.7-Max is current flagship",
        "rateSource": "https://www.alibabacloud.com/help/en/model-studio/model-pricing",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "qwen",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "qwen",
                "tier": "standard",
                "cost": "low",
                "input": 0.4,
                "output": 2.4,
                "bestFor": "cost-effective coding and multilingual tasks (catch-all, Qwen3.5-Plus DashScope rates)"
            },
            {
                "prefix": "qwen3.7-max",
                "tier": "reasoning",
                "cost": "low",
                "input": 1.25,
                "output": 3.75,
                "notes": "$1.25/$3.75 is a 50% promotional rate; standard list is $2.50/$7.50. International (Singapore) Model Studio endpoint; regional prices differ.",
                "bestFor": "current Qwen flagship (Qwen3.7-Max)",
                "inputTokenSoftLimit": 1000000
            },
            {
                "prefix": "qwen3.6-plus",
                "tier": "reasoning",
                "cost": "low",
                "input": 0.33,
                "output": 1.95,
                "notes": "rates from third-party aggregators (OpenRouter); official DashScope international pricing page does not list a dedicated qwen3.6-plus row.",
                "bestFor": "prior Qwen flagship (Qwen3.6-Plus, released 2026-03-30); 1M context",
                "inputTokenSoftLimit": 1000000
            },
            {
                "prefix": "qwen3.5-plus",
                "tier": "standard",
                "cost": "low",
                "input": 0.4,
                "output": 2.4,
                "notes": "tiered: $0.40/$2.40 ≤256k, $1.20/$3.60 above.",
                "bestFor": "Qwen3.5-Plus, tiered by context length"
            }
        ]
    },
    {
        "provider": "zhipu",
        "naming": "glm-{major}[.{minor}] — GLM-5.2 is current flagship; GLM-5 is ~30% more expensive than GLM-4",
        "rateSource": "https://docs.z.ai/guides/overview/pricing",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "glm",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "glm",
                "tier": "standard",
                "cost": "medium",
                "input": 1,
                "output": 3.2,
                "cachedRead": 0.2,
                "bestFor": "general GLM/Zhipu family tasks (catch-all, GLM-5 rates)"
            },
            {
                "prefix": "glm-5",
                "cachedRead": 0.2,
                "bestFor": "bilingual Chinese/English coding (GLM-5); 200k context",
                "inputTokenSoftLimit": 200000
            },
            {
                "prefix": "glm-5.2",
                "input": 1.4,
                "output": 4.4,
                "cachedRead": 0.26,
                "bestFor": "current GLM flagship (GLM-5.2, released 2026-06); agentic coding, 200k context",
                "inputTokenSoftLimit": 200000
            },
            {
                "prefix": "glm-5.1",
                "input": 1.4,
                "output": 4.4,
                "cachedRead": 0.26,
                "bestFor": "prior GLM flagship (GLM-5.1, released 2026-04-07); agentic coding, 200k context",
                "inputTokenSoftLimit": 200000
            },
            {
                "prefix": "glm-4",
                "cost": "low",
                "input": 0.6,
                "output": 2.2,
                "bestFor": "bilingual tasks at lower cost than GLM-5"
            },
            {
                "prefix": "glm-4.7",
                "cost": "low",
                "input": 0.6,
                "output": 2.2,
                "cachedRead": 0.11,
                "bestFor": "current GLM-4-series flagship"
            }
        ]
    },
    {
        "provider": "moonshot",
        "naming": "kimi-k{generation}[.{minor}] | kimi-k{generation}-{variant}",
        "rateSource": "https://platform.kimi.ai/docs/pricing/chat",
        "rateLookupDate": "2026-07-20",
        "defaults": {
            "family": "kimi",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "kimi",
                "tier": "standard",
                "cost": "low",
                "input": 0.6,
                "output": 3,
                "cachedRead": 0.1,
                "bestFor": "long-context tasks and coding (catch-all, K2.5 rates)"
            },
            {
                "prefix": "kimi-k2.5",
                "tier": "standard",
                "cost": "low",
                "input": 0.6,
                "output": 3,
                "cachedRead": 0.1,
                "bestFor": "prior Kimi flagship with vision + thinking modes"
            },
            {
                "prefix": "kimi-k2.6",
                "tier": "reasoning",
                "cost": "low",
                "input": 0.95,
                "output": 4,
                "cachedRead": 0.16,
                "bestFor": "current Kimi flagship (released 2026-04-20); 256k context, thinking + non-thinking modes",
                "inputTokenSoftLimit": 256000
            },
            {
                "prefix": "kimi-k3",
                "tier": "reasoning",
                "cost": "medium",
                "input": 3,
                "output": 15,
                "cachedRead": 0.6,
                "bestFor": "current Kimi flagship (K3, released 2026-07-16); 2.8T-param open MoE (16/896 experts active), native multimodal, 1M context; competitive with Fable 5, outperforms Opus 4.8 / GPT-5.6",
                "inputTokenSoftLimit": 1000000
            }
        ]
    },
    {
        "provider": "cohere",
        "naming": "command-r[-plus] — plus variant is flagship",
        "rateSource": "https://cohere.com/pricing",
        "rateLookupDate": "2026-06-01",
        "defaults": {
            "family": "cohere",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "command",
                "tier": "standard",
                "cost": "low",
                "input": 0.15,
                "output": 0.6,
                "bestFor": "general Cohere tasks (catch-all, Command R rates)"
            },
            {
                "prefix": "command-r",
                "tier": "standard",
                "cost": "low",
                "input": 0.15,
                "output": 0.6,
                "bestFor": "RAG workflows and grounded generation (current Command R)"
            },
            {
                "prefix": "command-r-plus",
                "tier": "standard",
                "cost": "medium",
                "input": 2.5,
                "output": 10,
                "bestFor": "higher-quality RAG and tool use (Command R+)"
            },
            {
                "prefix": "command-r7b",
                "tier": "standard",
                "cost": "low",
                "input": 0.037,
                "output": 0.15,
                "bestFor": "cheapest Cohere tier for high-throughput tasks"
            },
            {
                "prefix": "command-a",
                "tier": "reasoning",
                "cost": "medium",
                "input": 2.5,
                "output": 10,
                "bestFor": "Cohere flagship reasoning model"
            }
        ]
    },
    {
        "provider": "minimax",
        "naming": "MiniMax-M{major}[.{minor}][-highspeed] — case-sensitive prefix",
        "rateSource": "https://platform.minimax.io/docs/guides/pricing-paygo",
        "rateLookupDate": "2026-07-11",
        "defaults": {
            "family": "minimax",
            "supportsEffort": true,
            "inputTokenSoftLimit": 200000
        },
        "profiles": [
            {
                "prefix": "MiniMax-M2",
                "tier": "standard",
                "cost": "low",
                "input": 0.3,
                "output": 1.2,
                "bestFor": "well-scoped coding and agent loops where cost matters",
                "avoidFor": "highest-stakes ambiguous work that needs top-tier judgment"
            },
            {
                "prefix": "MiniMax-M2.7",
                "tier": "reasoning",
                "cost": "low",
                "input": 0.3,
                "output": 1.2,
                "cachedRead": 0.06,
                "bestFor": "prior MiniMax flagship (M2 generation); same rates as M2 with stronger reasoning"
            },
            {
                "prefix": "MiniMax-M3",
                "tier": "reasoning",
                "cost": "low",
                "input": 0.3,
                "output": 1.2,
                "cachedRead": 0.06,
                "notes": "tiered: $0.30/$1.20 ≤512k input tokens, $1.20/$4.80 above. Cost meter assumes ≤512k tier — long-context calls may underestimate. Toggleable thinking, same rates both modes.",
                "bestFor": "current MiniMax flagship (released 2026-06-01); 1M context sparse-attention reasoning at M2 prices",
                "inputTokenSoftLimit": 512000
            }
        ]
    },
    {
        "provider": "microsoft",
        "naming": "phi-{major}[-{variant}]",
        "rateSource": "https://azure.microsoft.com/en-us/pricing/details/cognitive-services/openai-service/",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "phi",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "phi",
                "tier": "standard",
                "cost": "low",
                "bestFor": "efficient open-weight Microsoft models"
            }
        ]
    },
    {
        "provider": "google-open",
        "naming": "gemma-{major}[.{minor}] — open-weight Gemma family",
        "rateSource": "https://ai.google.dev/gemma",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "gemma",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "gemma",
                "tier": "standard",
                "cost": "low",
                "bestFor": "lightweight open-weight Google models"
            }
        ]
    },
    {
        "provider": "yi-01",
        "naming": "yi-{tier}[-{version}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "yi",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "yi",
                "tier": "standard",
                "cost": "low",
                "bestFor": "bilingual Chinese/English open-weight models"
            }
        ]
    },
    {
        "provider": "perplexity",
        "naming": "sonar[-{variant}] — Perplexity's online-grounded models; per-request search fees not metered",
        "rateSource": "https://docs.perplexity.ai/docs/getting-started/pricing",
        "rateLookupDate": "2026-06-01",
        "defaults": {
            "family": "sonar",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "sonar",
                "tier": "standard",
                "cost": "low",
                "input": 1,
                "output": 1,
                "notes": "Perplexity also bills per-request search fees ($5–$12 per 1k requests) which token-based metering does not capture.",
                "bestFor": "online-grounded search-augmented tasks"
            },
            {
                "prefix": "sonar-pro",
                "tier": "standard",
                "cost": "medium",
                "input": 3,
                "output": 15,
                "bestFor": "higher-quality grounded search with citations; 200k context",
                "inputTokenSoftLimit": 200000
            },
            {
                "prefix": "sonar-reasoning",
                "tier": "reasoning",
                "cost": "medium",
                "input": 2,
                "output": 8,
                "bestFor": "reasoning over live web results (sonar-reasoning-pro / sonar-deep-research rates)"
            }
        ]
    },
    {
        "provider": "amazon",
        "naming": "nova-{tier}[-v{version}] — Amazon Bedrock native models (wire ids like amazon.nova-pro-v1:0)",
        "rateSource": "https://aws.amazon.com/bedrock/pricing/",
        "rateLookupDate": "2026-06-01",
        "defaults": {
            "family": "nova",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "nova",
                "tier": "standard",
                "cost": "low",
                "input": 0.8,
                "output": 3.2,
                "bestFor": "Amazon Bedrock native multimodal models (catch-all, Nova Pro rates)"
            },
            {
                "prefix": "nova-micro",
                "input": 0.035,
                "output": 0.14,
                "bestFor": "cheapest Nova tier, text-only"
            },
            {
                "prefix": "nova-lite",
                "input": 0.06,
                "output": 0.24,
                "bestFor": "low-cost multimodal Nova; 300k context",
                "inputTokenSoftLimit": 300000
            },
            {
                "prefix": "nova-pro",
                "input": 0.8,
                "output": 3.2,
                "bestFor": "balanced Nova multimodal; 300k context",
                "inputTokenSoftLimit": 300000
            },
            {
                "prefix": "nova-premier",
                "tier": "reasoning",
                "cost": "medium",
                "input": 2.5,
                "output": 12.5,
                "notes": "rates from secondary trackers; AWS pricing pages defer to console — re-confirm before relying on exact figures.",
                "bestFor": "Nova flagship for complex reasoning; ~1M context",
                "inputTokenSoftLimit": 1000000
            }
        ]
    },
    {
        "provider": "ai21",
        "naming": "jamba-{major}[.{minor}] — Mamba-transformer hybrid",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "jamba",
            "supportsEffort": false,
            "inputTokenSoftLimit": 256000
        },
        "profiles": [
            {
                "prefix": "jamba",
                "tier": "standard",
                "cost": "low",
                "bestFor": "long-context Mamba-transformer hybrid"
            }
        ]
    },
    {
        "provider": "ibm",
        "naming": "granite-{major}[.{minor}]-{variant}",
        "rateSource": "https://www.ibm.com/granite",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "granite",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "granite",
                "tier": "standard",
                "cost": "low",
                "bestFor": "enterprise open-weight IBM models"
            }
        ]
    },
    {
        "provider": "nvidia",
        "naming": "nemotron-{major}[-{variant}]",
        "rateSource": "https://build.nvidia.com/explore/discover",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "nemotron",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "nemotron",
                "tier": "standard",
                "cost": "low",
                "bestFor": "NVIDIA-optimized open-weight models"
            }
        ]
    },
    {
        "provider": "databricks",
        "naming": "dbrx[-{variant}]",
        "rateSource": "https://www.databricks.com/dbrx",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "dbrx",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "dbrx",
                "tier": "standard",
                "cost": "low",
                "bestFor": "mixture-of-experts enterprise models"
            }
        ]
    },
    {
        "provider": "snowflake",
        "naming": "arctic[-{variant}]",
        "rateSource": "https://www.snowflake.com/en/blog/snowflake-arctic-open-efficient-foundation-language-models-snowflake/",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "arctic",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "arctic",
                "tier": "standard",
                "cost": "low",
                "bestFor": "Dense-MoE hybrid Snowflake models"
            }
        ]
    },
    {
        "provider": "reka",
        "naming": "reka-{tier}[-{version}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "reka",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "reka",
                "tier": "standard",
                "cost": "low",
                "bestFor": "multimodal frontier models"
            }
        ]
    },
    {
        "provider": "ai2",
        "naming": "olmo-{major}[.{minor}][-{variant}] — fully open (data + code + weights)",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "olmo",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "olmo",
                "tier": "standard",
                "cost": "low",
                "bestFor": "fully open research models from AI2"
            }
        ]
    },
    {
        "provider": "nous-research",
        "naming": "hermes-{major}[.{minor}][-{variant}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "hermes",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "hermes",
                "tier": "standard",
                "cost": "low",
                "bestFor": "fine-tuned open-weight reasoning models"
            }
        ]
    },
    {
        "provider": "microsoft-wizardlm",
        "naming": "wizardlm-{major}[.{minor}][-{variant}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "wizardlm",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "wizardlm",
                "tier": "standard",
                "cost": "low",
                "bestFor": "instruction-following open-weight models"
            }
        ]
    },
    {
        "provider": "bigcode",
        "naming": "starcoder{generation}[-{variant}]",
        "rateSource": "https://huggingface.co/bigcode",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "starcoder",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "starcoder",
                "tier": "standard",
                "cost": "low",
                "bestFor": "code-generation open-weight models"
            }
        ]
    },
    {
        "provider": "cognitive-computations",
        "naming": "dolphin-{major}[.{minor}][-{variant}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "dolphin",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "dolphin",
                "tier": "standard",
                "cost": "low",
                "bestFor": "uncensored open-weight models"
            }
        ]
    },
    {
        "provider": "openchat",
        "naming": "openchat-{major}[.{minor}][-{variant}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "openchat",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "openchat",
                "tier": "standard",
                "cost": "low",
                "bestFor": "open-source chat-tuned models"
            }
        ]
    },
    {
        "provider": "lmsys",
        "naming": "vicuna-{major}[b][-{variant}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "vicuna",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "vicuna",
                "tier": "standard",
                "cost": "low",
                "bestFor": "LLaMA-finetuned chat models from LMSYS"
            }
        ]
    },
    {
        "provider": "shanghai-ai-lab",
        "naming": "internlm{major}[.{minor}][-{variant}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "internlm",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "internlm",
                "tier": "standard",
                "cost": "low",
                "bestFor": "multilingual Chinese/English open-weight models"
            }
        ]
    },
    {
        "provider": "baichuan",
        "naming": "baichuan{generation}[-{variant}]",
        "rateLookupDate": "2026-04-29",
        "defaults": {
            "family": "baichuan",
            "supportsEffort": false,
            "inputTokenSoftLimit": 128000
        },
        "profiles": [
            {
                "prefix": "baichuan",
                "tier": "standard",
                "cost": "low",
                "bestFor": "Chinese-language open-weight models"
            }
        ]
    }
]