From e7b15ac48ec66ff1bb01bb6161fd37ac94735ef9 Mon Sep 17 00:00:00 2001
From: Robert Putt <rob@puttfamily.co.uk>
Date: Mon, 13 Jul 2026 19:35:09 +0100
Subject: [PATCH] Add Nscale provider

---
 providers/nscale/logo.svg                     |  3 +++
 .../Qwen/Qwen2.5-Coder-32B-Instruct.toml      | 23 ++++++++++++++++
 .../Qwen/Qwen2.5-Coder-3B-Instruct.toml       | 23 ++++++++++++++++
 .../Qwen/Qwen2.5-Coder-7B-Instruct.toml       | 23 ++++++++++++++++
 .../Qwen/Qwen3-235B-A22B-Instruct-2507.toml   | 16 +++++++++++
 .../nscale/models/Qwen/Qwen3-235B-A22B.toml   |  7 +++++
 .../models/Qwen/Qwen3-4B-Instruct-2507.toml   | 23 ++++++++++++++++
 .../models/Qwen/Qwen3-4B-Thinking-2507.toml   | 24 +++++++++++++++++
 .../DeepSeek-R1-Distill-Llama-8B.toml         | 27 +++++++++++++++++++
 .../DeepSeek-R1-Distill-Qwen-14B.toml         | 27 +++++++++++++++++++
 .../DeepSeek-R1-Distill-Qwen-7B.toml          | 27 +++++++++++++++++++
 .../meta-llama/Llama-3.1-8B-Instruct.toml     | 24 +++++++++++++++++
 .../Llama-3.2-11B-Vision-Instruct.toml        | 24 +++++++++++++++++
 .../Llama-4-Scout-17B-16E-Instruct.toml       | 12 +++++++++
 .../models/mistralai/Devstral-Small-2505.toml | 24 +++++++++++++++++
 .../Mixtral-8x22B-Instruct-v0.1.toml          | 24 +++++++++++++++++
 .../nscale/models/openai/gpt-oss-120b.toml    |  9 +++++++
 .../nscale/models/openai/gpt-oss-20b.toml     |  9 +++++++
 providers/nscale/provider.toml                |  8 ++++++
 19 files changed, 357 insertions(+)
 create mode 100644 providers/nscale/logo.svg
 create mode 100644 providers/nscale/models/Qwen/Qwen2.5-Coder-32B-Instruct.toml
 create mode 100644 providers/nscale/models/Qwen/Qwen2.5-Coder-3B-Instruct.toml
 create mode 100644 providers/nscale/models/Qwen/Qwen2.5-Coder-7B-Instruct.toml
 create mode 100644 providers/nscale/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml
 create mode 100644 providers/nscale/models/Qwen/Qwen3-235B-A22B.toml
 create mode 100644 providers/nscale/models/Qwen/Qwen3-4B-Instruct-2507.toml
 create mode 100644 providers/nscale/models/Qwen/Qwen3-4B-Thinking-2507.toml
 create mode 100644 providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Llama-8B.toml
 create mode 100644 providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B.toml
 create mode 100644 providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B.toml
 create mode 100644 providers/nscale/models/meta-llama/Llama-3.1-8B-Instruct.toml
 create mode 100644 providers/nscale/models/meta-llama/Llama-3.2-11B-Vision-Instruct.toml
 create mode 100644 providers/nscale/models/meta-llama/Llama-4-Scout-17B-16E-Instruct.toml
 create mode 100644 providers/nscale/models/mistralai/Devstral-Small-2505.toml
 create mode 100644 providers/nscale/models/mistralai/Mixtral-8x22B-Instruct-v0.1.toml
 create mode 100644 providers/nscale/models/openai/gpt-oss-120b.toml
 create mode 100644 providers/nscale/models/openai/gpt-oss-20b.toml
 create mode 100644 providers/nscale/provider.toml

diff --git a/providers/nscale/logo.svg b/providers/nscale/logo.svg
new file mode 100644
index 0000000000..cd6f9ba1e3
--- /dev/null
+++ b/providers/nscale/logo.svg
@@ -0,0 +1,3 @@
+<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24" fill="currentColor">
+  <path d="M4 4h4v10.34L16 4h4v16h-4V9.66L8 20H4V4z"/>
+</svg>
diff --git a/providers/nscale/models/Qwen/Qwen2.5-Coder-32B-Instruct.toml b/providers/nscale/models/Qwen/Qwen2.5-Coder-32B-Instruct.toml
new file mode 100644
index 0000000000..4ec4dcbc09
--- /dev/null
+++ b/providers/nscale/models/Qwen/Qwen2.5-Coder-32B-Instruct.toml
@@ -0,0 +1,23 @@
+name = "Qwen 2.5 Coder 32B Instruct"
+description = "Large coding assistant for complex programming, refactoring, and code understanding"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+release_date = "2024-11-12"
+last_updated = "2024-11-12"
+open_weights = true
+
+[cost]
+input = 0.06
+output = 0.20
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/Qwen/Qwen2.5-Coder-3B-Instruct.toml b/providers/nscale/models/Qwen/Qwen2.5-Coder-3B-Instruct.toml
new file mode 100644
index 0000000000..0e38d54e58
--- /dev/null
+++ b/providers/nscale/models/Qwen/Qwen2.5-Coder-3B-Instruct.toml
@@ -0,0 +1,23 @@
+name = "Qwen 2.5 Coder 3B Instruct"
+description = "Compact coding assistant for code completion and generation"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+release_date = "2024-11-12"
+last_updated = "2024-11-12"
+open_weights = true
+
+[cost]
+input = 0.01
+output = 0.03
+
+[limit]
+context = 32_768
+input = 30_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/Qwen/Qwen2.5-Coder-7B-Instruct.toml b/providers/nscale/models/Qwen/Qwen2.5-Coder-7B-Instruct.toml
new file mode 100644
index 0000000000..72cfea05a4
--- /dev/null
+++ b/providers/nscale/models/Qwen/Qwen2.5-Coder-7B-Instruct.toml
@@ -0,0 +1,23 @@
+name = "Qwen 2.5 Coder 7B Instruct"
+description = "Mid-size coding assistant for code completion, generation, and understanding"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+release_date = "2024-11-12"
+last_updated = "2024-11-12"
+open_weights = true
+
+[cost]
+input = 0.01
+output = 0.03
+
+[limit]
+context = 32_768
+input = 30_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml b/providers/nscale/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml
new file mode 100644
index 0000000000..3818b84d2d
--- /dev/null
+++ b/providers/nscale/models/Qwen/Qwen3-235B-A22B-Instruct-2507.toml
@@ -0,0 +1,16 @@
+base_model = "alibaba/qwen3-235b-a22b"
+
+name = "Qwen3 235B A22B Instruct 2507"
+description = "Large Mixture-of-Experts instruction model for general-purpose chat and tool use"
+release_date = "2025-07-18"
+last_updated = "2025-07-18"
+reasoning = false
+structured_output = true
+
+[cost]
+input = 0.20
+output = 0.60
+
+[limit]
+input = 128_000
+output = 8_192
diff --git a/providers/nscale/models/Qwen/Qwen3-235B-A22B.toml b/providers/nscale/models/Qwen/Qwen3-235B-A22B.toml
new file mode 100644
index 0000000000..69a53d2c68
--- /dev/null
+++ b/providers/nscale/models/Qwen/Qwen3-235B-A22B.toml
@@ -0,0 +1,7 @@
+base_model = "alibaba/qwen3-235b-a22b"
+
+reasoning_options = []
+
+[cost]
+input = 0.20
+output = 0.60
diff --git a/providers/nscale/models/Qwen/Qwen3-4B-Instruct-2507.toml b/providers/nscale/models/Qwen/Qwen3-4B-Instruct-2507.toml
new file mode 100644
index 0000000000..cae128a681
--- /dev/null
+++ b/providers/nscale/models/Qwen/Qwen3-4B-Instruct-2507.toml
@@ -0,0 +1,23 @@
+name = "Qwen3 4B Instruct 2507"
+description = "Compact instruction model for lightweight chat and generation tasks"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+release_date = "2025-07-18"
+last_updated = "2025-07-18"
+open_weights = true
+
+[cost]
+input = 0.01
+output = 0.03
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/Qwen/Qwen3-4B-Thinking-2507.toml b/providers/nscale/models/Qwen/Qwen3-4B-Thinking-2507.toml
new file mode 100644
index 0000000000..a044ad29a1
--- /dev/null
+++ b/providers/nscale/models/Qwen/Qwen3-4B-Thinking-2507.toml
@@ -0,0 +1,24 @@
+name = "Qwen3 4B Thinking 2507"
+description = "Compact reasoning model with multi-step thinking capabilities"
+attachment = false
+reasoning = true
+reasoning_options = []
+tool_call = true
+structured_output = true
+temperature = true
+release_date = "2025-07-18"
+last_updated = "2025-07-18"
+open_weights = true
+
+[cost]
+input = 0.01
+output = 0.03
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Llama-8B.toml b/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Llama-8B.toml
new file mode 100644
index 0000000000..d37558f36d
--- /dev/null
+++ b/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Llama-8B.toml
@@ -0,0 +1,27 @@
+name = "DeepSeek R1 Distill Llama 8B"
+description = "Compact distilled reasoning model based on Llama architecture"
+attachment = false
+reasoning = true
+reasoning_options = []
+tool_call = false
+structured_output = false
+temperature = true
+release_date = "2025-01-20"
+last_updated = "2025-01-20"
+open_weights = true
+
+[cost]
+input = 0.05
+output = 0.05
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[interleaved]
+field = "reasoning_content"
diff --git a/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B.toml b/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B.toml
new file mode 100644
index 0000000000..198c9ecbab
--- /dev/null
+++ b/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-14B.toml
@@ -0,0 +1,27 @@
+name = "DeepSeek R1 Distill Qwen 14B"
+description = "Mid-size distilled reasoning model based on Qwen architecture"
+attachment = false
+reasoning = true
+reasoning_options = []
+tool_call = false
+structured_output = false
+temperature = true
+release_date = "2025-01-20"
+last_updated = "2025-01-20"
+open_weights = true
+
+[cost]
+input = 0.20
+output = 0.20
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[interleaved]
+field = "reasoning_content"
diff --git a/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B.toml b/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B.toml
new file mode 100644
index 0000000000..5a6a644294
--- /dev/null
+++ b/providers/nscale/models/deepseek-ai/DeepSeek-R1-Distill-Qwen-7B.toml
@@ -0,0 +1,27 @@
+name = "DeepSeek R1 Distill Qwen 7B"
+description = "Compact distilled reasoning model based on Qwen architecture optimized for math"
+attachment = false
+reasoning = true
+reasoning_options = []
+tool_call = false
+structured_output = false
+temperature = true
+release_date = "2025-01-20"
+last_updated = "2025-01-20"
+open_weights = true
+
+[cost]
+input = 0.15
+output = 0.15
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
+
+[interleaved]
+field = "reasoning_content"
diff --git a/providers/nscale/models/meta-llama/Llama-3.1-8B-Instruct.toml b/providers/nscale/models/meta-llama/Llama-3.1-8B-Instruct.toml
new file mode 100644
index 0000000000..5a537644dd
--- /dev/null
+++ b/providers/nscale/models/meta-llama/Llama-3.1-8B-Instruct.toml
@@ -0,0 +1,24 @@
+name = "Llama 3.1 8B Instruct"
+description = "Compact open Llama model for general chat and instruction following"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+knowledge = "2023-12"
+release_date = "2024-07-23"
+last_updated = "2024-07-23"
+open_weights = true
+
+[cost]
+input = 0.06
+output = 0.06
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/meta-llama/Llama-3.2-11B-Vision-Instruct.toml b/providers/nscale/models/meta-llama/Llama-3.2-11B-Vision-Instruct.toml
new file mode 100644
index 0000000000..1c34baf6d0
--- /dev/null
+++ b/providers/nscale/models/meta-llama/Llama-3.2-11B-Vision-Instruct.toml
@@ -0,0 +1,24 @@
+name = "Llama 3.2 11B Vision Instruct"
+description = "Multimodal vision-language model for image understanding and text generation"
+attachment = true
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+knowledge = "2023-12"
+release_date = "2024-09-25"
+last_updated = "2024-09-25"
+open_weights = true
+
+[cost]
+input = 0.06
+output = 0.06
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
+
+[modalities]
+input = ["text", "image"]
+output = ["text"]
diff --git a/providers/nscale/models/meta-llama/Llama-4-Scout-17B-16E-Instruct.toml b/providers/nscale/models/meta-llama/Llama-4-Scout-17B-16E-Instruct.toml
new file mode 100644
index 0000000000..fc1278e634
--- /dev/null
+++ b/providers/nscale/models/meta-llama/Llama-4-Scout-17B-16E-Instruct.toml
@@ -0,0 +1,12 @@
+base_model = "meta/llama-4-scout-17b-instruct"
+
+structured_output = true
+
+[cost]
+input = 0.09
+output = 0.29
+
+[limit]
+context = 131_072
+input = 128_000
+output = 8_192
diff --git a/providers/nscale/models/mistralai/Devstral-Small-2505.toml b/providers/nscale/models/mistralai/Devstral-Small-2505.toml
new file mode 100644
index 0000000000..d2d0d88176
--- /dev/null
+++ b/providers/nscale/models/mistralai/Devstral-Small-2505.toml
@@ -0,0 +1,24 @@
+name = "Devstral Small 2505"
+description = "Mistral coding agent model for repository tasks and software engineering workflows"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+knowledge = "2025-05"
+release_date = "2025-05-21"
+last_updated = "2025-05-21"
+open_weights = true
+
+[cost]
+input = 0.10
+output = 0.30
+
+[limit]
+context = 128_000
+input = 124_000
+output = 8_192
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/mistralai/Mixtral-8x22B-Instruct-v0.1.toml b/providers/nscale/models/mistralai/Mixtral-8x22B-Instruct-v0.1.toml
new file mode 100644
index 0000000000..34302abc40
--- /dev/null
+++ b/providers/nscale/models/mistralai/Mixtral-8x22B-Instruct-v0.1.toml
@@ -0,0 +1,24 @@
+name = "Mixtral 8x22B Instruct v0.1"
+description = "Large Mixture-of-Experts model for general-purpose instruction following"
+attachment = false
+reasoning = false
+tool_call = true
+structured_output = true
+temperature = true
+knowledge = "2024-04"
+release_date = "2024-04-17"
+last_updated = "2024-04-17"
+open_weights = true
+
+[cost]
+input = 1.20
+output = 1.20
+
+[limit]
+context = 65_536
+input = 63_000
+output = 4_096
+
+[modalities]
+input = ["text"]
+output = ["text"]
diff --git a/providers/nscale/models/openai/gpt-oss-120b.toml b/providers/nscale/models/openai/gpt-oss-120b.toml
new file mode 100644
index 0000000000..d892b65a92
--- /dev/null
+++ b/providers/nscale/models/openai/gpt-oss-120b.toml
@@ -0,0 +1,9 @@
+# reasoning_effort "low"|"medium"|"high" supported via OpenAI-compatible chat completions
+# https://docs.nscale.com/api-reference/inference/create-chat-completion (accessed 2026-07-13)
+base_model = "openai/gpt-oss-120b"
+
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[cost]
+input = 0.10
+output = 0.40
diff --git a/providers/nscale/models/openai/gpt-oss-20b.toml b/providers/nscale/models/openai/gpt-oss-20b.toml
new file mode 100644
index 0000000000..c1ebb1f4f0
--- /dev/null
+++ b/providers/nscale/models/openai/gpt-oss-20b.toml
@@ -0,0 +1,9 @@
+# reasoning_effort "low"|"medium"|"high" supported via OpenAI-compatible chat completions
+# https://docs.nscale.com/api-reference/inference/create-chat-completion (accessed 2026-07-13)
+base_model = "openai/gpt-oss-20b"
+
+reasoning_options = [{ type = "effort", values = ["low", "medium", "high"] }]
+
+[cost]
+input = 0.05
+output = 0.20
diff --git a/providers/nscale/provider.toml b/providers/nscale/provider.toml
new file mode 100644
index 0000000000..7b96f6d017
--- /dev/null
+++ b/providers/nscale/provider.toml
@@ -0,0 +1,8 @@
+# Pricing and model details: https://nscale.com/serverless (accessed 2026-07-13)
+# API docs: https://docs.nscale.com/docs/use-cases/chat (accessed 2026-07-13)
+# Base URL and auth: https://docs.nscale.com/docs/use-cases/chat (accessed 2026-07-13)
+name = "Nscale"
+env = ["NSCALE_API_KEY"]
+npm = "@ai-sdk/openai-compatible"
+api = "https://inference.api.nscale.com/v1"
+doc = "https://docs.nscale.com/docs/ai-services/models"