Skip to content
Open
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
43 changes: 43 additions & 0 deletions providers/aigrow/models/claude-haiku-5.5.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,43 @@
# Prices: https://aigrow.gg/ai-api, price table "US dollars per million input / output tokens", row "Claude Haiku 5.5", column "With AIGROW credit" (read 2026-11-16): input $0.04, output $0.20, cache read $0.01, cache write $0.05. Below the table: a request with more than 100,000 prompt tokens, cache reads and writes included, bills all its tokens at $0.20 input, $0.05 cache read, $0.25 cache write and $1.00 output. These are the USD a buyer pays per 1M tokens, copied with no conversion. Cache writes bill at 1.25x the input rate.
# APIs: Anthropic Messages (POST /v1/messages) and OpenAI Chat Completions (POST /v1/chat/completions) on the same key: https://aigrow.gg/ai-api/guide
# Effort: output_config.effort (Messages) and reasoning_effort (Chat Completions); levels as on providers/anthropic; each level returned 200 on this host on both APIs (2026-11-16).
# Cache: cache_control breakpoints work on Messages (a write on the first send, a read on the second), but a breakpoint that asks for the 1-hour TTL is written to the 5-minute cache: the usage split reported it as ephemeral_5m_input_tokens (probed 2026-11-16). So only 5m is listed, and a write bills at 1.25x the input rate. Breakpoints, targets and minimum copied from providers/anthropic.
# Mid-conversation flags and tool strict/search are left unset: not tested on this host.
# Structured output: a JSON schema in response_format (Chat Completions) and in output_config.format (Messages) returned bare JSON when the prompt asked for those fields, and plain prose when it did not (probed 2026-11-16), so the schema is not enforced on either API and the flag is false.
# Input: a PNG image was read on both APIs. A PDF was not: a file part (Chat Completions) and a document block (Messages) were each counted as about 2,100 prompt tokens, but on two attempts per API the model answered that no PDF was attached (probed 2026-11-16), so PDF is left out.
# Tools: a function tool returned a tool call with the right arguments on Chat Completions with tool_choice "auto" and on Messages with tool_choice auto (probed 2026-11-16).
# Streaming: with stream_options.include_usage the last chunk before [DONE] carried the usage on Chat Completions, and message_delta carried it on Messages (probed 2026-11-16).
schema = 2
base_model = "anthropic/claude-haiku-5-5"

[capabilities]
reasoning = { supported = true, effort = ["low", "medium", "high", "xhigh", "max"] }
structured_output = false
stream_usage = true
cache = { explicit = { ttl = ["5m"], max_breakpoints = 4, targets = ["tools", "system", "messages"], min_tokens = 512 } }

[modalities]
input = ["text", "image/jpeg", "image/png", "image/gif", "image/webp"]
output = ["text"]

[api.messages]
base_url = "https://aigrow.gg/v1"
sdk = { aisdk = "@ai-sdk/anthropic", opencodeai = "@opencode/ai/providers/anthropic-compatible" }

[api.chat-completions]

[cost.text.input]
price = 0.04
cache = { read = 0.01, write = { "5m" = 0.05 } }

[[cost.text.input.overrides]]
when = { context = { gt = 100_000 } }
price = 0.2
cache = { read = 0.05, write = { "5m" = 0.25 } }

[cost.text.output]
price = 0.2

[[cost.text.output.overrides]]
when = { context = { gt = 100_000 } }
price = 1
31 changes: 31 additions & 0 deletions providers/aigrow/models/gemini-3.8-flash.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
# Prices: https://aigrow.gg/ai-api, price table "US dollars per million input / output tokens", row "Gemini 3.8 Flash", column "With AIGROW credit" (read 2026-11-16): input $0.375, output $1.875, cache read $0.0375, cache write $0.4688. These are the USD a buyer pays per 1M tokens, copied with no conversion, except the cache write: the page rounds it to four decimals, and the file keeps the rate it rounds, 1.25x the $0.375 input rate ($0.46875). One flat rate, with no date-based change on this host.
# API: OpenAI Chat Completions only (POST /v1/chat/completions); the Messages endpoint does not serve this model (404, probed 2026-11-16).
# Effort: reasoning_effort; levels as on providers/google; each level returned 200 with reasoning tokens reported, and the reasoning comes back in reasoning_content; an assistant turn sent back with its reasoning_content was accepted (200) (2026-11-16). No default is stated: this host's default was not confirmed.
# Cache: implicit; minimum copied from providers/google. Three identical 9,721-token prompts reported no cached tokens on the first two sends and 7,674 on the third (probed 2026-11-16). Explicit caching is left out: not tested on this host.
# Output: one request returns at most 65,536 output tokens, thinking included, the limit copied from providers/google (2026-11-16).
# Structured output: a JSON schema response_format returned bare JSON, also on a prompt that did not ask for JSON (probed 2026-11-16).
# Input: a PNG image and a PDF file part were read (probed 2026-11-16). Audio and video are left out because they were not tested on this host.
# Tools: a function tool returned a tool call with the right arguments on Chat Completions with tool_choice "auto"; the legacy functions field returns 400 (probed 2026-11-16).
# Streaming: with stream_options.include_usage the last chunk before [DONE] carried the usage (probed 2026-11-16).
schema = 2
base_model = "google/gemini-3.8-flash"

[capabilities]
reasoning = { supported = true, effort = ["low", "medium", "high"] }
structured_output = true
stream_usage = true
cache = { implicit = { min_tokens = 4_096 } }

[modalities]
input = ["text", "image/png", "image/jpeg", "image/webp", "image/heic", "image/heif", "application/pdf"]
output = ["text"]

[api.chat-completions]
compatibility = { reasoning_field = "reasoning_content" }

[cost.text.input]
price = 0.375
cache = { read = 0.0375, write = 0.46875 }

[cost.text.output]
price = 1.875
31 changes: 31 additions & 0 deletions providers/aigrow/models/gemini-4-hydrogen.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,31 @@
# Prices: https://aigrow.gg/ai-api, price table "US dollars per million input / output tokens", row "Gemini 4 Hydrogen", column "With AIGROW credit" (read 2026-11-16): input $0.75, output $3.75, cache read $0.0375, cache write $0.9375. These are the USD a buyer pays per 1M tokens, copied with no conversion. One flat rate at every context length; cache writes bill at 1.25x the input rate.
# API: OpenAI Chat Completions only (POST /v1/chat/completions); the Messages endpoint does not serve this model (404, probed 2026-11-16).
# Effort: reasoning_effort low, medium and high each returned 200 with reasoning tokens reported, and the reasoning comes back in reasoning_content; an assistant turn sent back with its reasoning_content was accepted (200) (2026-11-16). xhigh returned 200 as well, but this host runs it as high, so it is left out. No default is stated: this host's default was not confirmed.
# Cache: implicit; minimum copied from providers/google. Three identical 9,754-token prompts reported no cached tokens, then 7,700 and 7,700 (probed 2026-11-16).
# Output: one request returns at most 65,536 output tokens, thinking included, the limit copied from providers/google (2026-11-16).
# Structured output: a JSON schema response_format returned bare JSON, also on a prompt that did not ask for JSON (probed 2026-11-16).
# Input: a PNG image and a PDF file part were read (probed 2026-11-16). Audio and video are left out because they were not tested on this host.
# Tools: a function tool returned a tool call with the right arguments on Chat Completions with tool_choice "auto"; the legacy functions field returns 400 (probed 2026-11-16).
# Streaming: with stream_options.include_usage the last chunk before [DONE] carried the usage (probed 2026-11-16).
schema = 2
base_model = "google/gemini-4-hydrogen"

[capabilities]
reasoning = { supported = true, effort = ["low", "medium", "high"] }
structured_output = true
stream_usage = true
cache = { implicit = { min_tokens = 4_096 } }

[modalities]
input = ["text", "image/png", "image/jpeg", "image/webp", "image/heic", "image/heif", "application/pdf"]
output = ["text"]

[api.chat-completions]
compatibility = { reasoning_field = "reasoning_content" }

[cost.text.input]
price = 0.75
cache = { read = 0.0375, write = 0.9375 }

[cost.text.output]
price = 3.75
21 changes: 21 additions & 0 deletions providers/aigrow/models/gpt-6-luna.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,21 @@
# Prices: https://aigrow.gg/ai-api, price table "US dollars per million input / output tokens", row "GPT-6 Luna", column "With AIGROW credit" (read 2026-11-16): input $0.025, output $0.125, cache read $0.01, cache write $0.0313. These are the USD a buyer pays per 1M tokens, copied with no conversion, except the cache write: the page rounds it to four decimals, and the file keeps the rate it rounds, 1.25x the $0.025 input rate ($0.03125). One flat rate at every context length (no long-context tier).
# API: OpenAI Chat Completions (POST /v1/chat/completions); there is no Responses API. /v1/messages also answered this model (200, 2026-11-16), but nothing else was probed there, so it is not listed.
# Effort: reasoning_effort; levels as on providers/openai; each level returned 200 on this host, with reasoning tokens reported from medium up (2026-11-16). A JSON schema response_format, a PNG image and a PDF file part all worked, and the schema held on a prompt that did not ask for JSON (probed 2026-11-16).
# Cache: automatic prefix caching; minimum copied from providers/openai. A second identical 3,498-token prompt reported 3,328 cached tokens (probed 2026-11-16).
# Streaming: with stream_options.include_usage the last chunk before [DONE] carried the usage (probed 2026-11-16).
# Tools: custom (freeform) tools do not reach the model on this host, so freeform is left out. Probed 2026-11-16 on Chat Completions with one custom tool per format (text, a regex grammar ZZ[0-9]{3}, the same grammar in lark): with tool_choice "required" each request returned 400 "Tool choice 'required' must be specified with 'tools' parameter.", and with no tool_choice each returned 200 with no tool call and the model answering that the tool is not available. Function tools are forwarded: a function tool returned a tool call with the right arguments under tool_choice "required" and "auto" (probed 2026-11-16).
schema = 2
base_model = "openai/gpt-6-luna"

[capabilities]
reasoning = { supported = true, effort = ["none", "low", "medium", "high", "xhigh", "max"] }
structured_output = true
stream_usage = true
cache = { implicit = { min_tokens = 1_024 } }

[cost.text.input]
price = 0.025
cache = { read = 0.01, write = 0.03125 }

[cost.text.output]
price = 0.125
29 changes: 29 additions & 0 deletions providers/aigrow/models/kimi-k3.1-pro.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,29 @@
# Prices: https://aigrow.gg/ai-api, price table "US dollars per million input / output tokens", row "Kimi K3.1 Pro", column "With AIGROW credit" (read 2026-11-16): input $1.00, output $5.00. These are the USD a buyer pays per 1M tokens, copied with no conversion. One flat rate at every context length. The row lists no cache rates; it reads "No prompt cache: repeated context bills as input".
# API: OpenAI Chat Completions only (POST /v1/chat/completions); the Messages endpoint does not serve this model (404, probed 2026-11-16).
# Effort: reasoning_effort; levels as on providers/moonshotai; low, high and max each returned 200 (2026-11-16). medium returned 200 too, but it is not one of Moonshot's levels, so it is left out. The reasoning comes back in reasoning_content, while the usage reports 0 reasoning tokens, and an assistant turn sent back with its reasoning_content was accepted (200). No default is stated: this host's default was not confirmed.
# Cache: none. A second identical 3,550-token prompt reported 0 cached tokens (probed 2026-11-16), as the price page says.
# Structured output: a JSON schema response_format returned bare JSON when the prompt asked for those fields, and plain prose when it did not (probed 2026-11-16), so the schema is not enforced and the flag is false.
# Input: a PNG image was read (probed 2026-11-16). Video is left out because it was not tested on this host. A PDF file part, which providers/moonshotai does not list either, was not read.
# Tools: a function tool returned a tool call with the right arguments on Chat Completions with tool_choice "auto" (probed 2026-11-16).
# Streaming: with stream_options.include_usage the last chunk before [DONE] carried the usage (probed 2026-11-16).
schema = 2
base_model = "moonshotai/kimi-k3.1-pro"

[capabilities]
reasoning = { supported = true, effort = ["low", "high", "max"] }
structured_output = false
stream_usage = true
cache = false

[modalities]
input = ["text", "image/jpeg", "image/png", "image/gif", "image/webp", "image/bmp", "image/heic", "image/heif"]
output = ["text"]

[api.chat-completions]
compatibility = { reasoning_field = "reasoning_content" }

[cost.text.input]
price = 1

[cost.text.output]
price = 5
30 changes: 30 additions & 0 deletions providers/aigrow/models/qwen3.8-max.toml
Original file line number Diff line number Diff line change
@@ -0,0 +1,30 @@
# Prices: https://aigrow.gg/ai-api, price table "US dollars per million input / output tokens", row "Qwen3.8 Max", column "With AIGROW credit" (read 2026-11-16): input $1.00, output $3.00, cache read $0.125, cache write $1.25. These are the USD a buyer pays per 1M tokens, copied with no conversion. One flat rate at every context length; cache writes bill at 1.25x the input rate.
# API: OpenAI Chat Completions only (POST /v1/chat/completions); the Messages endpoint does not serve this model (404, probed 2026-11-16).
# Effort: reasoning_effort; levels as on providers/alibaba; low, medium and xhigh each returned 200 with reasoning tokens reported, and so did high, an alias of xhigh on providers/alibaba (2026-11-16). The reasoning comes back in reasoning_content, and an assistant turn sent back with its reasoning_content was accepted (200). No default is stated: this host's default was not confirmed.
# Cache: implicit prefix caching; minimum copied from providers/alibaba. A second identical 4,111-token prompt reported 4,096 cached tokens (probed 2026-11-16). Explicit cache_control breakpoints are left out: not tested on this host.
# Structured output: a JSON schema response_format returned bare JSON, also on a prompt that did not ask for JSON (probed 2026-11-16).
# Input: a PNG image was read. A PDF file part returned 502 on both of two attempts (2026-11-16), so PDF is left out; video is left out because it was not tested on this host.
# Tools: a function tool returned a tool call with the right arguments on Chat Completions with tool_choice "auto" (probed 2026-11-16).
# Streaming: with stream_options.include_usage the last chunk before [DONE] carried the usage (probed 2026-11-16).
schema = 2
base_model = "alibaba/qwen3.8-max"

[capabilities]
reasoning = { supported = true, effort = ["low", "medium", "xhigh"] }
structured_output = true
stream_usage = true
cache = { implicit = { min_tokens = 1_024 } }

[modalities]
input = ["text", "image/bmp", "image/jpeg", "image/png", "image/tiff", "image/webp", "image/heic"]
output = ["text"]

[api.chat-completions]
compatibility = { reasoning_field = "reasoning_content" }

[cost.text.input]
price = 1
cache = { read = 0.125, write = 1.25 }

[cost.text.output]
price = 3
27 changes: 16 additions & 11 deletions providers/aigrow/provider.toml
Original file line number Diff line number Diff line change
@@ -1,14 +1,19 @@
# AIGROW sells prepaid API credit for Claude, GPT, Gemini and Grok models on one
# endpoint that speaks OpenAI Chat Completions (POST /v1/chat/completions) and
# Anthropic Messages (POST /v1/messages). Keys start with `ag_` and are sent as
# `Authorization: Bearer` or `x-api-key`. GET /v1/models lists the served ids
# (it needs a key). Setup: https://aigrow.gg/ai-api/guide
# Prices: https://aigrow.gg/ai-api (read 2026-11-12). The figure on that page is
# headed "USD per million tokens" and lists, for each model, the USD a buyer pays
# per 1M tokens. The model files copy those figures with no conversion, and each
# one names its row. Each model has one flat rate at every context length. A
# prompt-cache write bills at 1.25x the input rate, and at 2x on the Claude
# models, where every write goes to the 1-hour cache.
# AIGROW sells prepaid API credit for Claude, GPT, Gemini, Grok, Qwen and Kimi
# models on one endpoint that speaks OpenAI Chat Completions
# (POST /v1/chat/completions) and Anthropic Messages (POST /v1/messages). Keys
# start with `ag_` and are sent as `Authorization: Bearer` or `x-api-key`.
# GET /v1/models lists the served ids (it needs a key). Setup:
# https://aigrow.gg/ai-api/guide
# Prices: https://aigrow.gg/ai-api (read 2026-11-16). The price table on that
# page reads "US dollars per million input / output tokens" (files read before
# 2026-11-16 call it the figure "USD per million tokens") and lists, for each
# model, the USD a buyer pays per 1M tokens. The model files copy those figures
# with no conversion, and each one names its row. Each model has one flat rate
# at every context length, except Claude Haiku 5.5: a request with more than
# 100,000 prompt tokens bills at its higher tier. A prompt-cache write bills at
# 1.25x the input rate, and at 2x on Claude Opus 5.5, Sonnet 5.5, Fable 5.1 and
# Fable 5.2, where every write goes to the 1-hour cache. Kimi K3.1 Pro reports
# no prompt-cache reads, so repeated context bills as input.
name = "AIGROW"
npm = "@ai-sdk/openai-compatible"
opencodeai = "@opencode/ai/providers/openai-compatible"
Expand Down
Loading