diff --git a/labs/vispark/lab.toml b/labs/vispark/lab.toml new file mode 100644 index 00000000000..3aebe3fe5da --- /dev/null +++ b/labs/vispark/lab.toml @@ -0,0 +1 @@ +description = "Vispark builds India-centered foundational multimodal intelligence (Vision) with 1M-token context, served via Vispark Lab's API." diff --git a/labs/vispark/logo.svg b/labs/vispark/logo.svg new file mode 100644 index 00000000000..2348c5a2f7f --- /dev/null +++ b/labs/vispark/logo.svg @@ -0,0 +1,3 @@ + + + diff --git a/models/vispark/vision-large.toml b/models/vispark/vision-large.toml new file mode 100644 index 00000000000..0c850177b0c --- /dev/null +++ b/models/vispark/vision-large.toml @@ -0,0 +1,26 @@ +# Lab metadata for Vispark Vision Large. +# Provider-agnostic facts about the model Vispark built. +# Served first-party by Vispark Lab via OpenAI-compatible API: https://api.lab.vispark.in/v1 +# Model list: GET https://api.lab.vispark.in/v1/models returns vispark/vision-large +# (verified live 2026-09-13; context_length 1000000, max_output_length 65536 per entry). +# Chat: POST https://api.lab.vispark.in/v1/chat/completions. +# Release 2024-05-15: live endpoint returns created = 1715731200 (= 2024-05-15) per model, +# Last updated 2026-09: current lineup as verified live 2026-09-13 (month precision). +name = "Vision Large" +description = "Most capable Vision model for complex reasoning, detailed media analysis, and structured output over a 1M-token context window." +release_date = "2024-05-15" +last_updated = "2026-09" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "audio", "video", "pdf"] +output = ["text"] diff --git a/models/vispark/vision-medium.toml b/models/vispark/vision-medium.toml new file mode 100644 index 00000000000..89b8b000c7f --- /dev/null +++ b/models/vispark/vision-medium.toml @@ -0,0 +1,26 @@ +# Lab metadata for Vispark Vision Medium. +# Provider-agnostic facts about the model Vispark built. +# Served first-party by Vispark Lab via OpenAI-compatible API: https://api.lab.vispark.in/v1 +# Model list: GET https://api.lab.vispark.in/v1/models returns vispark/vision-medium +# (verified live 2026-09-13; context_length 1000000, max_output_length 65536 per entry). +# Chat: POST https://api.lab.vispark.in/v1/chat/completions. +# Release 2024-05-15: live endpoint returns created = 1715731200 (= 2024-05-15) per model, +# Last updated 2026-09: current lineup as verified live 2026-09-13 (month precision). +name = "Vision Medium" +description = "Balanced multimodal model pairing a 1M-token context window with deeper reasoning for analysis, content creation, and tool use across text, image, audio, video, and PDF inputs." +release_date = "2024-05-15" +last_updated = "2026-09" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "audio", "video", "pdf"] +output = ["text"] diff --git a/models/vispark/vision-small.toml b/models/vispark/vision-small.toml new file mode 100644 index 00000000000..2dce89d0c4f --- /dev/null +++ b/models/vispark/vision-small.toml @@ -0,0 +1,26 @@ +# Lab metadata for Vispark Vision Small. +# Provider-agnostic facts about the model Vispark built. +# Served first-party by Vispark Lab via OpenAI-compatible API: https://api.lab.vispark.in/v1 +# Model list: GET https://api.lab.vispark.in/v1/models returns vispark/vision-small +# (verified live 2026-09-13; context_length 1000000, max_output_length 65536 per entry). +# Chat: POST https://api.lab.vispark.in/v1/chat/completions. +# Release 2024-05-15: live endpoint returns created = 1715731200 (= 2024-05-15) per model, +# Last updated 2026-09: current lineup as verified live 2026-09-13 (month precision). +name = "Vision Small" +description = "Fast, low-cost multimodal model for understanding text, images, audio, video, and PDFs, with tool calling and a 1M-token context window." +release_date = "2024-05-15" +last_updated = "2026-09" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "audio", "video", "pdf"] +output = ["text"] diff --git a/providers/vispark/logo.svg b/providers/vispark/logo.svg new file mode 100644 index 00000000000..2348c5a2f7f --- /dev/null +++ b/providers/vispark/logo.svg @@ -0,0 +1,3 @@ + + + diff --git a/providers/vispark/models/vispark/vision-large.toml b/providers/vispark/models/vispark/vision-large.toml new file mode 100644 index 00000000000..9238a65cb89 --- /dev/null +++ b/providers/vispark/models/vispark/vision-large.toml @@ -0,0 +1,36 @@ +# Vispark: Vision Large — first-party host (provider IS the lab, full inline allowed). +# API model ID: vispark/vision-large — GET https://api.lab.vispark.in/v1/models (verified live 2026-09-13). +# Cost source, USD/MTok, used verbatim: the provider's own models endpoint publishes +# pricing.prompt = "0.00000737" and pricing.completion = "0.00002211" USD per token +# (per-token USD strings) = input 7.37, output 22.11 per MTok. +# Reasoning: always-on, no caller control — accepted params are only +# temperature/max_tokens/top_p/top_k/stop/seed/frequency_penalty/presence_penalty/tools/ +# tool_choice/response_format — no reasoning_effort/thinking/enable_thinking/budget field; +# Capabilities: temperature, top_p/top_k/stop/seed, tools, response_format json_object/json_schema, +# multimodal input text+image+audio+video+pdf (base64/URL), text output, streaming + usage. +# Limits: context 1_000_000 (context_length), output 65_536 (max_output_length) — same endpoint. +# Release 2024-05-15: live endpoint returns created = 1715731200 (= 2024-05-15) per model, +# Last updated 2026-09: current lineup as verified live 2026-09-13 (month precision). +name = "Vision Large" +description = "Most capable Vision model for complex reasoning, detailed media analysis, and structured output over a 1M-token context window." +release_date = "2024-05-15" +last_updated = "2026-09" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false +reasoning_options = [] + +[cost] +input = 7.37 +output = 22.11 + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "audio", "video", "pdf"] +output = ["text"] diff --git a/providers/vispark/models/vispark/vision-medium.toml b/providers/vispark/models/vispark/vision-medium.toml new file mode 100644 index 00000000000..08a3a70579d --- /dev/null +++ b/providers/vispark/models/vispark/vision-medium.toml @@ -0,0 +1,36 @@ +# Vispark: Vision Medium — first-party host (provider IS the lab, full inline allowed). +# API model ID: vispark/vision-medium (default model) — GET https://api.lab.vispark.in/v1/models (verified live 2026-09-13). +# Cost source, USD/MTok, used verbatim: the provider's own models endpoint publishes +# pricing.prompt = "0.00000421" and pricing.completion = "0.00001263" USD per token +# (per-token USD strings) = input 4.21, output 12.63 per MTok. +# Reasoning: always-on, no caller control — accepted params are only +# temperature/max_tokens/top_p/top_k/stop/seed/frequency_penalty/presence_penalty/tools/ +# tool_choice/response_format — no reasoning_effort/thinking/enable_thinking/budget field; +# Capabilities: temperature, top_p/top_k/stop/seed, tools, response_format json_object/json_schema, +# multimodal input text+image+audio+video+pdf (base64/URL), text output, streaming + usage. +# Limits: context 1_000_000 (context_length), output 65_536 (max_output_length) — same endpoint. +# Release 2024-05-15: live endpoint returns created = 1715731200 (= 2024-05-15) per model, +# Last updated 2026-09: current lineup as verified live 2026-09-13 (month precision). +name = "Vision Medium" +description = "Balanced multimodal model pairing a 1M-token context window with deeper reasoning for analysis, content creation, and tool use across text, image, audio, video, and PDF inputs." +release_date = "2024-05-15" +last_updated = "2026-09" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false +reasoning_options = [] + +[cost] +input = 4.21 +output = 12.63 + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "audio", "video", "pdf"] +output = ["text"] diff --git a/providers/vispark/models/vispark/vision-small.toml b/providers/vispark/models/vispark/vision-small.toml new file mode 100644 index 00000000000..c93c050eb49 --- /dev/null +++ b/providers/vispark/models/vispark/vision-small.toml @@ -0,0 +1,36 @@ +# Vispark: Vision Small — first-party host (provider IS the lab, full inline allowed). +# API model ID: vispark/vision-small — GET https://api.lab.vispark.in/v1/models (verified live 2026-09-13). +# Cost source, USD/MTok, used verbatim: the provider's own models endpoint publishes +# pricing.prompt = "0.00000105" and pricing.completion = "0.00000316" USD per token +# (per-token USD strings) = input 1.05, output 3.16 per MTok. +# Reasoning: always-on, no caller control — accepted params are only +# temperature/max_tokens/top_p/top_k/stop/seed/frequency_penalty/presence_penalty/tools/ +# tool_choice/response_format — no reasoning_effort/thinking/enable_thinking/budget field; +# Capabilities: temperature, top_p/top_k/stop/seed, tools, response_format json_object/json_schema, +# multimodal input text+image+audio+video+pdf (base64/URL), text output, streaming + usage. +# Limits: context 1_000_000 (context_length), output 65_536 (max_output_length) — same endpoint. +# Release 2024-05-15: live endpoint returns created = 1715731200 (= 2024-05-15) per model, +# Last updated 2026-09: current lineup as verified live 2026-09-13 (month precision). +name = "Vision Small" +description = "Fast, low-cost multimodal model for understanding text, images, audio, video, and PDFs, with tool calling and a 1M-token context window." +release_date = "2024-05-15" +last_updated = "2026-09" +attachment = true +reasoning = true +temperature = true +tool_call = true +structured_output = true +open_weights = false +reasoning_options = [] + +[cost] +input = 1.05 +output = 3.16 + +[limit] +context = 1_000_000 +output = 65_536 + +[modalities] +input = ["text", "image", "audio", "video", "pdf"] +output = ["text"] diff --git a/providers/vispark/provider.toml b/providers/vispark/provider.toml new file mode 100644 index 00000000000..ac7e1634846 --- /dev/null +++ b/providers/vispark/provider.toml @@ -0,0 +1,11 @@ +# Vispark Lab — first-party host for Vispark Vision models. +# API: https://api.lab.vispark.in/v1 +# Endpoints: GET /v1/models, POST /v1/chat/completions, POST /v1/completions +# Auth: X-API-Key header or Authorization: Bearer (keys like vl_...) +# Platform: https://lab.vispark.in/ — get API key: https://lab.vispark.in/profile#api +# Vision docs: https://lab.vispark.in/#vision — company: https://vispark.in +name = "Vispark" +npm = "@ai-sdk/openai-compatible" +env = ["VISPARK_LAB_API_KEY"] +api = "https://api.lab.vispark.in/v1" +doc = "https://lab.vispark.in/#vision"