Skip to content

:hedgehog: PostHog is the leading platform for building self-driving products. Our developer tools – AI observability, analytics, session replay, flags, experiments, error tracking, logs, and more – capture all the context agents need to diagnose problems, uncover opportunities, and ship fixes. Steer it all from Slack, web, desktop, or the MCP.

18.1 KB
Raw Blame
// AUTO-GENERATED by products/tasks/scripts/model_catalog_projection.py from
// products/tasks/backend/model_catalog.py.
// Regenerate with `hogli build:projections`. Do not edit.
//
// The single definition of how a task agent run is configured: harness,
// runtime adapter, model, and reasoning effort, plus what each model
// costs. The backend validates a run against this same data, so a picker
// built on this file can only offer a selection the API will accept.

export type RuntimeAdapter = 'claude' | 'codex'

export type Runtime = 'acp' | 'pi'

export interface RuntimeOption {
    runtime: Runtime
    /** Absent for Pi, which has no adapter. */
    runtimeAdapter?: RuntimeAdapter
    label: string
}

/** What a harness picker shows. The runtime says which agent program runs the
    task; the adapter says which vendor protocol ACP speaks. A picker shows one
    flat list, so the two choices become one set of entries here. */
export const RUNTIME_OPTIONS: readonly RuntimeOption[] = [
    {
        runtime: 'acp',
        runtimeAdapter: 'claude',
        label: 'Claude Code',
    },
    {
        runtime: 'acp',
        runtimeAdapter: 'codex',
        label: 'Codex',
    },
    {
        runtime: 'pi',
        label: 'Pi',
    },
]

/** Thinking depths, shallowest first. */
export type ReasoningEffort = 'low' | 'medium' | 'high' | 'xhigh' | 'max' | 'ultracode'

export const REASONING_EFFORTS: readonly ReasoningEffort[] = ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode']

/** What a picker calls each depth. */
export const REASONING_EFFORT_LABELS: Record<ReasoningEffort, string> = {
    low: 'Low',
    medium: 'Medium',
    high: 'High',
    xhigh: 'Extra high',
    max: 'Max',
    ultracode: 'Ultracode',
}

/** A value for each runtime adapter. */
export type ByRuntimeAdapter<T> = Record<RuntimeAdapter, T>

/** Efforts per adapter, absent where the adapter has no fallback. */
type FallbackEfforts = Partial<ByRuntimeAdapter<readonly ReasoningEffort[]>>

export interface LongContextCost {
    aboveInputTokens: number
    inputPerMtok: number
    outputPerMtok: number
}

export interface ModelCost {
    inputPerMtok: number
    outputPerMtok: number
    longContext?: LongContextCost
}

export interface CatalogModel {
    id: string
    runtimeAdapter: RuntimeAdapter
    /** Empty for a model with no effort control: render no dropdown. */
    reasoningEfforts: readonly ReasoningEffort[]
    /**
     * What a picker shows. Resolved when this file is generated, so every
     * surface names a model the same way without carrying a formatter.
     */
    label: string
    /**
     * Feature flag a person needs before a picker offers this model. Absent
     * means generally available. Governs display only — the server decides
     * whether a run may use it.
     */
    accessFlag?: string
    /**
     * Base list price in US dollars per million tokens. Absent for a model no
     * public price list covers, which a picker shows with no cost at all.
     */
    cost?: ModelCost
    /**
     * Per-token cost against Claude Sonnet 5, ready to render: `2.5×`,
     * `≈0.55×`, or `1× base`. Prefixed when input and output rates diverge
     * enough that one number flatters either. Absent whenever `cost` is.
     */
    costMultiplier?: string
    /**
     * The rates behind the multiplier, ready to render. Absent whenever
     * `cost` is.
     */
    costSummary?: string
    /**
     * Runs with the 1M-token context window. Absent means it does not, and a
     * picker offers no window choice.
     */
    supports1MContext?: boolean
    /**
     * Runs in fast mode. Absent means it does not, and a picker offers no
     * fast-mode toggle.
     */
    supportsFastMode?: boolean
    /**
     * Superseded: no picker offers it, and a session already pinned to it
     * still runs and still reads its name and cost from here.
     */
    retired?: boolean
}

/** The model `1×` refers to. */
export const COST_BASELINE_MODEL = 'claude-sonnet-5'

export const COST_BASELINE_LABEL = 'Claude Sonnet 5'

/** Which vendor API each runtime adapter speaks. */
export const PROVIDER_BY_RUNTIME_ADAPTER: ByRuntimeAdapter<string> = {
    claude: 'anthropic',
    codex: 'openai',
}

export const RUNTIME_ADAPTERS: readonly RuntimeAdapter[] = ['claude', 'codex']

export const MODELS: readonly CatalogModel[] = [
    {
        id: '@cf/zai-org/glm-5.2',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['high', 'max'],
        label: 'GLM-5.2',
        cost: {
            inputPerMtok: 1.4,
            outputPerMtok: 4.4,
        },
        costMultiplier: '≈0.57×',
        costSummary: 'Input $1.40 · Output $4.40 per 1M tokens',
        retired: true,
    },
    {
        id: 'zai-org/glm-5.2',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['high', 'max'],
        label: 'GLM-5.2',
        cost: {
            inputPerMtok: 1.4,
            outputPerMtok: 4.4,
        },
        costMultiplier: '≈0.57×',
        costSummary: 'Input $1.40 · Output $4.40 per 1M tokens',
        retired: true,
    },
    {
        id: 'zai-org/glm-5.3',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['high', 'max'],
        label: 'GLM-5.3',
        cost: {
            inputPerMtok: 1.4,
            outputPerMtok: 4.4,
        },
        costMultiplier: '≈0.57×',
        costSummary: 'Input $1.40 · Output $4.40 per 1M tokens',
    },
    {
        id: 'zai-org/glm-5.3-flash',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['high', 'max'],
        label: 'GLM-5.3 Flash',
        cost: {
            inputPerMtok: 0.15,
            outputPerMtok: 0.5,
        },
        costMultiplier: '≈0.06×',
        costSummary: 'Input $0.15 · Output $0.50 per 1M tokens',
    },
    {
        id: 'moonshotai/kimi-k3',
        runtimeAdapter: 'claude',
        reasoningEfforts: [],
        label: 'Kimi K3',
        cost: {
            inputPerMtok: 3,
            outputPerMtok: 15,
        },
        costMultiplier: '1.5×',
        costSummary: 'Input $3 · Output $15 per 1M tokens',
    },
    {
        id: 'deepseek-ai/deepseek-v4-flash-0731',
        runtimeAdapter: 'claude',
        reasoningEfforts: [],
        label: 'DeepSeek V4 Flash',
        cost: {
            inputPerMtok: 0.13,
            outputPerMtok: 0.26,
        },
        costMultiplier: '≈0.05×',
        costSummary: 'Input $0.13 · Output $0.26 per 1M tokens',
    },
    {
        id: 'deepseek-ai/deepseek-v4.1-flash-1017-fast',
        runtimeAdapter: 'claude',
        reasoningEfforts: [],
        label: 'DeepSeek V4.1 Flash Fast',
        cost: {
            inputPerMtok: 0.6,
            outputPerMtok: 2.4,
        },
        costMultiplier: '≈0.27×',
        costSummary: 'Input $0.60 · Output $2.40 per 1M tokens',
    },
    {
        id: 'claude-opus-4-5',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high'],
        label: 'Claude Opus 4.5',
        cost: {
            inputPerMtok: 5,
            outputPerMtok: 25,
        },
        costMultiplier: '2.5×',
        costSummary: 'Input $5 · Output $25 per 1M tokens',
        retired: true,
    },
    {
        id: 'claude-opus-4-6',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'Claude Opus 4.6',
        cost: {
            inputPerMtok: 5,
            outputPerMtok: 25,
        },
        costMultiplier: '2.5×',
        costSummary: 'Input $5 · Output $25 per 1M tokens',
        retired: true,
    },
    {
        id: 'claude-opus-4-7',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Opus 4.7',
        cost: {
            inputPerMtok: 5,
            outputPerMtok: 25,
        },
        costMultiplier: '2.5×',
        costSummary: 'Input $5 · Output $25 per 1M tokens',
        supports1MContext: true,
        supportsFastMode: true,
        retired: true,
    },
    {
        id: 'claude-opus-4-8',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Opus 4.8',
        cost: {
            inputPerMtok: 5,
            outputPerMtok: 25,
        },
        costMultiplier: '2.5×',
        costSummary: 'Input $5 · Output $25 per 1M tokens',
        supports1MContext: true,
        supportsFastMode: true,
    },
    {
        id: 'claude-opus-5',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Opus 5',
        cost: {
            inputPerMtok: 5,
            outputPerMtok: 25,
        },
        costMultiplier: '2.5×',
        costSummary: 'Input $5 · Output $25 per 1M tokens',
        supports1MContext: true,
        supportsFastMode: true,
    },
    {
        id: 'claude-opus-5-5',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Opus 5.5',
        cost: {
            inputPerMtok: 4,
            outputPerMtok: 20,
        },
        costMultiplier: '2×',
        costSummary: 'Input $4 · Output $20 per 1M tokens',
        supports1MContext: true,
        supportsFastMode: true,
    },
    {
        id: 'claude-fable-5',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Fable 5',
        cost: {
            inputPerMtok: 10,
            outputPerMtok: 50,
        },
        costMultiplier: '5×',
        costSummary: 'Input $10 · Output $50 per 1M tokens',
        supports1MContext: true,
    },
    {
        id: 'claude-fable-5-1',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Fable 5.1',
        cost: {
            inputPerMtok: 10,
            outputPerMtok: 50,
        },
        costMultiplier: '5×',
        costSummary: 'Input $10 · Output $50 per 1M tokens',
        supports1MContext: true,
    },
    {
        id: 'claude-sonnet-5-5',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Sonnet 5.5',
        cost: {
            inputPerMtok: 2,
            outputPerMtok: 10,
        },
        costMultiplier: '1×',
        costSummary: 'Input $2 · Output $10 per 1M tokens',
        supports1MContext: true,
    },
    {
        id: 'claude-sonnet-5',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max', 'ultracode'],
        label: 'Claude Sonnet 5',
        cost: {
            inputPerMtok: 2,
            outputPerMtok: 10,
        },
        costMultiplier: '1×',
        costSummary: 'Input $2 · Output $10 per 1M tokens',
        supports1MContext: true,
    },
    {
        id: 'claude-sonnet-4-6',
        runtimeAdapter: 'claude',
        reasoningEfforts: ['low', 'medium', 'high'],
        label: 'Claude Sonnet 4.6',
        cost: {
            inputPerMtok: 3,
            outputPerMtok: 15,
        },
        costMultiplier: '1.5×',
        costSummary: 'Input $3 · Output $15 per 1M tokens',
        supports1MContext: true,
        retired: true,
    },
    {
        id: 'gpt-5',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high'],
        label: 'GPT-5',
    },
    {
        id: 'gpt-5.5',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh'],
        label: 'GPT-5.5',
        cost: {
            inputPerMtok: 5,
            outputPerMtok: 30,
        },
        costMultiplier: '≈2.8×',
        costSummary: 'Input $5 · Output $30 per 1M tokens',
    },
    {
        id: 'gpt-5.6-sol',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-5.6 Sol',
        cost: {
            inputPerMtok: 5,
            outputPerMtok: 30,
        },
        costMultiplier: '≈2.8×',
        costSummary: 'Input $5 · Output $30 per 1M tokens',
    },
    {
        id: 'gpt-5.6-terra',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-5.6 Terra',
        cost: {
            inputPerMtok: 2.5,
            outputPerMtok: 15,
        },
        costMultiplier: '≈1.4×',
        costSummary: 'Input $2.50 · Output $15 per 1M tokens',
    },
    {
        id: 'gpt-5.6-luna',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-5.6 Luna',
        cost: {
            inputPerMtok: 1,
            outputPerMtok: 6,
        },
        costMultiplier: '≈0.55×',
        costSummary: 'Input $1 · Output $6 per 1M tokens',
    },
    {
        id: 'gpt-6-astra',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-6 Astra',
        cost: {
            inputPerMtok: 10,
            outputPerMtok: 50,
        },
        costMultiplier: '5×',
        costSummary: 'Input $10 · Output $50 per 1M tokens',
    },
    {
        id: 'gpt-6-sol',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-6 Sol',
        cost: {
            inputPerMtok: 2,
            outputPerMtok: 10,
        },
        costMultiplier: '1×',
        costSummary: 'Input $2 · Output $10 per 1M tokens',
    },
    {
        id: 'gpt-6.1-sol',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-6.1 Sol',
        cost: {
            inputPerMtok: 2,
            outputPerMtok: 10,
            longContext: {
                aboveInputTokens: 272000,
                inputPerMtok: 4,
                outputPerMtok: 15,
            },
        },
        costMultiplier: '1× base',
        costSummary: 'Per 1M tokens: $2 input/$10 output to 272K; $4 input/$15 output above',
    },
    {
        id: 'gpt-6.1-terra',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-6.1 Terra',
        cost: {
            inputPerMtok: 1,
            outputPerMtok: 5,
            longContext: {
                aboveInputTokens: 272000,
                inputPerMtok: 2,
                outputPerMtok: 7.5,
            },
        },
        costMultiplier: '0.5× base',
        costSummary: 'Per 1M tokens: $1 input/$5 output to 272K; $2 input/$7.50 output above',
        retired: true,
    },
    {
        id: 'gpt-6-luna',
        runtimeAdapter: 'codex',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
        label: 'GPT-6 Luna',
        cost: {
            inputPerMtok: 0.1,
            outputPerMtok: 0.5,
        },
        costMultiplier: '0.05×',
        costSummary: 'Input $0.10 · Output $0.50 per 1M tokens',
    },
]

/** The model a run uses when it pins none. */
export const DEFAULT_MODEL_BY_RUNTIME_ADAPTER: ByRuntimeAdapter<string> = {
    claude: 'claude-sonnet-5-5',
    codex: 'gpt-6.1-sol',
}

export interface CapabilityNotch {
    model: string
    effort: ReasoningEffort
}

type CapabilityLadders = ByRuntimeAdapter<readonly CapabilityNotch[]>

/** The Faster → Smarter rungs each harness offers, cheapest first. Filter them
    against what the gateway serves before rendering, so a rung naming a retired
    model drops out instead of becoming a stop that fails on send. */
export const CAPABILITY_LADDER_BY_RUNTIME_ADAPTER: CapabilityLadders = {
    claude: [
        {
            model: 'claude-sonnet-5-5',
            effort: 'medium',
        },
        {
            model: 'claude-sonnet-5-5',
            effort: 'high',
        },
        {
            model: 'claude-opus-5-5',
            effort: 'medium',
        },
        {
            model: 'claude-opus-5-5',
            effort: 'xhigh',
        },
        {
            model: 'claude-fable-5-1',
            effort: 'max',
        },
    ],
    codex: [
        {
            model: 'gpt-6-luna',
            effort: 'low',
        },
        {
            model: 'gpt-6-sol',
            effort: 'low',
        },
        {
            model: 'gpt-6-sol',
            effort: 'medium',
        },
        {
            model: 'gpt-6-sol',
            effort: 'high',
        },
        {
            model: 'gpt-6-sol',
            effort: 'xhigh',
        },
        {
            model: 'gpt-6-astra',
            effort: 'max',
        },
    ],
}

export interface ModelFamily {
    runtimeAdapter: RuntimeAdapter
    prefix: string
    reasoningEfforts: readonly ReasoningEffort[]
}

// Depths a whole family exposes, used when no exact id matches. OpenAI ships
// vendor-flavoured variants the gateway never lists (`gpt-5.5-codex`), and a
// local session can drive one. The longest matching prefix wins, so the
// declaration order below is free.
export const FAMILY_REASONING_EFFORTS: readonly ModelFamily[] = [
    {
        runtimeAdapter: 'codex',
        prefix: 'gpt-6-astra',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
    },
    {
        runtimeAdapter: 'codex',
        prefix: 'gpt-6-sol',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
    },
    {
        runtimeAdapter: 'codex',
        prefix: 'gpt-6.1-sol',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
    },
    {
        runtimeAdapter: 'codex',
        prefix: 'gpt-6-luna',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
    },
    {
        runtimeAdapter: 'codex',
        prefix: 'gpt-5.6',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh', 'max'],
    },
    {
        runtimeAdapter: 'codex',
        prefix: 'gpt-5.5',
        reasoningEfforts: ['low', 'medium', 'high', 'xhigh'],
    },
]

// What a runtime accepts for a model neither an id nor a family claims.
// Codex passes any `gpt-*` identifier through, so an id the gateway starts
// serving before it is listed here still runs. Claude is absent on
// purpose: an unnamed model there yields no efforts, which is what makes
// validation reject it.
export const FALLBACK_REASONING_EFFORTS: FallbackEfforts = {
    codex: ['low', 'medium', 'high'],
}