diff --git a/skills/clawrouter/SKILL.md b/skills/clawrouter/SKILL.md index db972d4e..01149dcc 100644 --- a/skills/clawrouter/SKILL.md +++ b/skills/clawrouter/SKILL.md @@ -132,14 +132,14 @@ Routing is done locally by [`@blockrun/router-core`](https://github.com/BlockRun Curated primaries on the default `auto` profile: -- **SIMPLE** — factual lookups, greetings, translations → gemini-2.5-flash ($0.30/$2.50) -- **MEDIUM** — summaries, explanations, data extraction → kimi-k2.7 ($0.95/$4.00) -- **COMPLEX** — code generation, multi-step analysis → gemini-3.1-pro ($2/$12) -- **REASONING** — proofs, formal logic, multi-step math → grok-4-1-fast-reasoning ($0.20/$0.50) +- **SIMPLE** — factual lookups, greetings, translations → gemini-2.5-flash +- **MEDIUM** — summaries, explanations, data extraction → kimi-k2.7 +- **COMPLEX** — code generation, multi-step analysis → gemini-3.1-pro +- **REASONING** — proofs, formal logic, multi-step math → grok-4-1-fast-reasoning Turns that actually need their attached tools switch to agent-tuned tiers (gpt-4o-mini → kimi-k2.7 → claude-sonnet-4.6). `eco` opens on the free tier (nemotron-3.5-lightning); `premium` climbs to gpt-5.3-codex / claude-fable-5 / claude-sonnet-4.6. Full chains: [docs/routing-profiles.md](../../docs/routing-profiles.md). -Prices are per 1M input/output tokens. Per-tier +Per-tier savings percentages are deliberately not quoted here: the published figure is blended across a stated workload mix, and a per-tier number invites comparing it against a baseline nobody wrote down. See @@ -147,7 +147,34 @@ it against a baseline nobody wrote down. See ## Available Models -88 models including: gpt-6-astra [GPT-6 flagship, 1M ctx], gpt-6-sol, gpt-6.1-sol [refreshed Sol, $2/$10], gpt-6-luna [fast, $0.10/$0.50], gpt-5.6-terra [balanced, stable default], gpt-5.6-sol [flagship], gpt-5.6-luna [cost-efficient], gpt-5.6-sol/terra/luna-pro [pro reasoning tiers], gpt-5.5, gpt-5.5-pro [max compute], chat-latest [ChatGPT Instant], gpt-5.4, gpt-5.1, gpt-4o, o3, claude-fable-5.1, claude-fable-5, claude-opus-5.5 [newest Opus, $4/$20], claude-opus-5 [1M ctx], claude-opus-4.8, claude-opus-4.7, claude-opus-4.5, claude-sonnet-5.5, claude-sonnet-5, claude-sonnet-4.6, gemini-3.1-pro, gemini-3.6-flash, gemini-3.5-flash-lite, gemini-2.5-flash, deepseek-v4-pro, deepseek-chat, grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-build-0.1, glm-5.3 [Z.AI flagship, 1M ctx], glm-5.3-flash [natively multimodal, 1M ctx — $0.15/$0.50], glm-5.2, minimax-m3, kimi-k3 [1M ctx flagship], kimi-k2.7, qwen3.8-max [Qwen 3.8 flagship, 1M ctx], qwen3.7-max, qwen3.7-plus, qwen3.7-flash, mimo [Xiaomi MiMo-V2.5 Pro, 1M ctx], mistral-large-4 [Mistral flagship, 512K ctx, vision], hy4-preview [Tencent, 770B MoE], laguna-s-2.1 [Poolside, coding, $0.09/$0.18], and 7 free open-weight models (nemotron-3.5-lightning [1M ctx, thinking mode — the free default], nemotron-3-ultra-550b [550B/55B MoE, 1M ctx — the largest free model], nemotron-3-nano-omni-30b-a3b-reasoning [256K ctx], llama-3.2-11b-vision [Meta Llama], north-mini-code [Cohere, coding, sub-second], laguna-xs-2.1 [Poolside, coding, ~161 tok/s], muse-glimmer-30b [Meta Muse, reasoning]). The free tier is text-only — image turns route to a paid vision model, because the two free models the catalogs advertise as vision-capable fail a solid-colour probe on both chains and do so with an HTTP 200. +88 models including: gpt-6-astra [GPT-6 flagship, 1M ctx], gpt-6-sol, gpt-6.1-sol [refreshed Sol], gpt-6-luna [fast], gpt-5.6-terra [balanced, stable default], gpt-5.6-sol [flagship], gpt-5.6-luna [cost-efficient], gpt-5.6-sol/terra/luna-pro [pro reasoning tiers], gpt-5.5, gpt-5.5-pro [max compute], chat-latest [ChatGPT Instant], gpt-5.4, gpt-5.1, gpt-4o, o3, claude-fable-5.1, claude-fable-5, claude-opus-5.5 [newest Opus], claude-opus-5 [1M ctx], claude-opus-4.8, claude-opus-4.7, claude-opus-4.5, claude-sonnet-5.5, claude-sonnet-5, claude-sonnet-4.6, gemini-3.1-pro, gemini-3.6-flash, gemini-3.5-flash-lite, gemini-2.5-flash, deepseek-v4-pro, deepseek-chat, grok-4.7, grok-4.6, grok-4.5, grok-4.3, grok-build-0.1, glm-5.3 [Z.AI flagship, 1M ctx], glm-5.3-flash [natively multimodal, 1M ctx], glm-5.2, minimax-m3, kimi-k3 [1M ctx flagship], kimi-k2.7, qwen3.8-max [Qwen 3.8 flagship, 1M ctx], qwen3.7-max, qwen3.7-plus, qwen3.7-flash, mimo [Xiaomi MiMo-V2.5 Pro, 1M ctx], mistral-large-4 [Mistral flagship, 512K ctx, vision], hy4-preview [Tencent, 770B MoE], laguna-s-2.1 [Poolside, coding], and 7 free open-weight models (nemotron-3.5-lightning [1M ctx, thinking mode — the free default], nemotron-3-ultra-550b [550B/55B MoE, 1M ctx — the largest free model], nemotron-3-nano-omni-30b-a3b-reasoning [256K ctx], llama-3.2-11b-vision [Meta Llama], north-mini-code [Cohere, coding, sub-second], laguna-xs-2.1 [Poolside, coding, ~161 tok/s], muse-glimmer-30b [Meta Muse, reasoning]). The free tier is text-only — image turns route to a paid vision model, because the two free models the catalogs advertise as vision-capable fail a solid-colour probe on both chains and do so with an HTTP 200. + +## Model Prices — Look Them Up Live + +Model prices move (vendors reprice, promos end), so this skill carries no model price table. Read the live catalog before you budget a call: + +```bash +curl -s https://blockrun.ai/api/v1/models # Base +curl -s https://sol.blockrun.ai/api/v1/models # Solana +``` + +Each entry in `data[]` has an `id`, a `billing_mode` and a `pricing` object. All amounts are USD. + +| `billing_mode` | `pricing` fields | Estimated cost of one call | +| ---------------- | ------------------------------------------------------------------------------------------------------ | -------------------------------------------------------------------------------------------------------------- | +| `paid` (chat) | `input`, `output` per 1M tokens; optional `cache_read` / `cache_write` per 1M; optional `long_context` | `(input_tokens × input + output_tokens × output) / 1,000,000`, minimum $0.001, plus the $0.001 transaction fee | +| `free` | `input: 0`, `output: 0` | $0 — no payment | +| `per_image` | `per_image`, `transaction_fee` | `per_image × images + transaction_fee` | +| `per_second` | `per_second`, `per_call_surcharge`, `transaction_fee`, `default_duration_seconds` | `per_second × seconds + per_call_surcharge + transaction_fee` | +| `per_track` | `per_track`, `transaction_fee` | `per_track + transaction_fee` | +| `per_character` | `per_1k_chars`, `transaction_fee` | `per_1k_chars × characters / 1,000 + transaction_fee` | +| `per_generation` | `per_generation`, `transaction_fee` | `per_generation + transaction_fee` | + +- **Cached prompt tokens** bill at `cache_read` instead of `input` (and cache writes at `cache_write`) when the model lists them. +- **`long_context`** is a list of tiers. When the prompt is larger than a tier's `threshold` (or equal to it, if `inclusive` is true), that tier's `input` / `output` / `cache_*` rates apply to the **whole** request. +- Image size/quality and video resolution can raise the media price above the listed base rate. +- The figure above is an estimate. The exact amount is quoted in the HTTP 402 challenge before anything is signed, so a call never settles above its quote. +- Human-readable rate card: https://blockrun.ai/pricing ## Built-in Agent Tools @@ -168,11 +195,11 @@ Realtime prices and historical OHLC across every asset class. The agent should c ### Image & Video Generation -| Tool | Purpose | Price | -| --------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------------------------------------------- | -| `blockrun_image_generation` | 12 image models — GPT Image 1/2/2.5 (Flare, Sunburst), Nano Banana / 2 / Pro, Seedream 5 Pro, Grok Imagine / 2.0 / Pro, CogView-4 | $0.015–$0.15 / image | -| `blockrun_image_edit` | Edit / inpaint existing image (GPT Image 1/2, 2.5 Sunburst, Nano Banana; `model` param) | $0.02–$0.04 / image | -| `blockrun_video_generation` | Grok Imagine + ByteDance Seedance (1.5-pro / 2.0-fast / 2.0) at 720p with synced audio (t2v default), 5–10s, + OpenAI Sora 2 via Azure (`azure/sora-2`, 4/8/12s, t2v + i2v, rejects human faces in reference images). Seedance is token-priced upstream (~20,256 tokens/sec at the 720p+audio default — 2× the prior 480p rate); `image_url` (image-to-video) costs the same as t2v since 2026-06-01. Seedance 2.0 variants accept optional `real_face_asset_id` (`ta_…`) for BytePlus RealFace character-consistency — mutually exclusive with `image_url`. | $0.05/s (Grok); $0.10/s (Sora 2); Seedance ~$0.46–$1.49 per 5s clip | +| Tool | Purpose | Price | +| --------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | ---------------------------------------------------------------- | +| `blockrun_image_generation` | 12 image models — GPT Image 1/2/2.5 (Flare, Sunburst), Nano Banana / 2 / Pro, Seedream 5 Pro, Grok Imagine / 2.0 / Pro, CogView-4 | per model — see [Model Prices](#model-prices--look-them-up-live) | +| `blockrun_image_edit` | Edit / inpaint existing image (GPT Image 1/2, 2.5 Sunburst, Nano Banana; `model` param) | per model — see [Model Prices](#model-prices--look-them-up-live) | +| `blockrun_video_generation` | Grok Imagine + ByteDance Seedance (1.5-pro / 2.0-fast / 2.0) at 720p with synced audio (t2v default), 5–10s, + OpenAI Sora 2 via Azure (`azure/sora-2`, 4/8/12s, t2v + i2v, rejects human faces in reference images). Seedance is token-priced upstream (~20,256 tokens/sec at the 720p+audio default — 2× the prior 480p rate); `image_url` (image-to-video) costs the same as t2v since 2026-06-01. Seedance 2.0 variants accept optional `real_face_asset_id` (`ta_…`) for BytePlus RealFace character-consistency — mutually exclusive with `image_url`. | per model — see [Model Prices](#model-prices--look-them-up-live) | ### Phone & Voice (Twilio + Bland.ai) diff --git a/skills/imagegen/SKILL.md b/skills/imagegen/SKILL.md index a94baed4..1f963407 100644 --- a/skills/imagegen/SKILL.md +++ b/skills/imagegen/SKILL.md @@ -55,20 +55,22 @@ Display inline: `![generated image](http://localhost:8402/images/abc123.png)` ### Model Selection -| Alias | Full ID | Price | Sizes | Best for | -| ------------------ | ------------------------------- | ------------ | ------------------------------- | ------------------------------------------------------------------------------------------------------------------ | -| `nano-banana` | `google/nano-banana` | $0.05 | 1024×1024 | Default — fast, cheap, good quality | -| `banana-2` | `google/nano-banana-2` | $0.09 | 1024×1024 | Gemini 3.1 Flash imagegen — sharper than nano-banana at 1K | -| `banana-pro` | `google/nano-banana-pro` | $0.10–$0.15 | 1024×1024, 2048×2048, 4096×4096 | High-res, large format | -| `gpt-image` | `openai/gpt-image-1` | $0.02–$0.04 | 1024×1024, 1536×1024, 1024×1536 | Budget option; supports editing | -| `gpt-image-2` | `openai/gpt-image-2` | $0.06–$0.12 | 1024×1024, 1536×1024, 1024×1536 | Photorealistic, reasoning-driven, text rendering (slow — proxy polls up to 5min); legacy `dalle` alias routes here | -| `flare` | `openai/gpt-image-2.5-flare` | $0.28–$0.56 | 1024×1024, 1536×1024, 1024×1536 | GPT Image 2.5, fast — top OpenAI quality, caller can pick `quality` at a flat price | -| `sunburst` | `openai/gpt-image-2.5-sunburst` | $0.28–$0.56 | 1024×1024, 1536×1024, 1024×1536 | GPT Image 2.5, precision — best for high-fidelity edits | -| `seedream` | `bytedance/seedream-5-pro` | $0.045–$0.09 | up to 2848×1600 / 2304×1728 | Flagship quality, reference-image support | -| `grok-imagine` | `xai/grok-imagine-image` | $0.02 | 1024×1024 | xAI Grok image style | -| `grok-imagine-2` | `xai/grok-imagine-image-2.0` | $0.04 | 1024×1024 | Grok Imagine 2.0 — between grok-imagine and pro | -| `grok-imagine-pro` | `xai/grok-imagine-image-pro` | $0.07 | 1024×1024 | Grok high-quality | -| `cogview` | `zai/cogview-4` | $0.015–$0.02 | 512×512 to 1440×1440 | Cheapest — Zhipu CogView | +| Alias | Full ID | Sizes | Best for | +| ------------------ | ------------------------------- | ------------------------------- | ------------------------------------------------------------------------------------------------------------------ | +| `nano-banana` | `google/nano-banana` | 1024×1024 | Default — fast, cheap, good quality | +| `banana-2` | `google/nano-banana-2` | 1024×1024 | Gemini 3.1 Flash imagegen — sharper than nano-banana at 1K | +| `banana-pro` | `google/nano-banana-pro` | 1024×1024, 2048×2048, 4096×4096 | High-res, large format | +| `gpt-image` | `openai/gpt-image-1` | 1024×1024, 1536×1024, 1024×1536 | Budget option; supports editing | +| `gpt-image-2` | `openai/gpt-image-2` | 1024×1024, 1536×1024, 1024×1536 | Photorealistic, reasoning-driven, text rendering (slow — proxy polls up to 5min); legacy `dalle` alias routes here | +| `flare` | `openai/gpt-image-2.5-flare` | 1024×1024, 1536×1024, 1024×1536 | GPT Image 2.5, fast — top OpenAI quality, caller can pick `quality` at a flat price | +| `sunburst` | `openai/gpt-image-2.5-sunburst` | 1024×1024, 1536×1024, 1024×1536 | GPT Image 2.5, precision — best for high-fidelity edits | +| `seedream` | `bytedance/seedream-5-pro` | up to 2848×1600 / 2304×1728 | Flagship quality, reference-image support | +| `grok-imagine` | `xai/grok-imagine-image` | 1024×1024 | xAI Grok image style | +| `grok-imagine-2` | `xai/grok-imagine-image-2.0` | 1024×1024 | Grok Imagine 2.0 — between grok-imagine and pro | +| `grok-imagine-pro` | `xai/grok-imagine-image-pro` | 1024×1024 | Grok high-quality | +| `cogview` | `zai/cogview-4` | 512×512 to 1440×1440 | Cheapest — Zhipu CogView | + +**Prices:** per image, and they change — look the model up live instead of trusting a number here. `GET https://blockrun.ai/api/v1/models` (Solana: `https://sol.blockrun.ai/api/v1/models`), find the entry whose `id` is the Full ID above, and read `pricing.per_image` plus `pricing.transaction_fee`. Larger sizes or higher `quality` can cost more than the listed base rate; the exact amount is quoted in the HTTP 402 challenge before anything is signed. **Choosing a model:** @@ -119,7 +121,7 @@ Response is identical to generation: } ``` -**Supported models for editing:** `openai/gpt-image-1` (default, $0.02), `openai/gpt-image-2` ($0.06), `openai/gpt-image-2.5-sunburst` ($0.28), `google/nano-banana` ($0.05), `google/nano-banana-2` ($0.09), `google/nano-banana-pro` ($0.10). `mask` works with the OpenAI models only. `openai/gpt-image-2.5-flare` cannot edit. +**Supported models for editing:** `openai/gpt-image-1` (default), `openai/gpt-image-2`, `openai/gpt-image-2.5-sunburst`, `google/nano-banana`, `google/nano-banana-2`, `google/nano-banana-pro`. `mask` works with the OpenAI models only. `openai/gpt-image-2.5-flare` cannot edit. ---