diff --git a/api-reference/models/gpt-oss-20b.mdx b/api-reference/models/gpt-oss-20b.mdx new file mode 100644 index 0000000..bb0ef8c --- /dev/null +++ b/api-reference/models/gpt-oss-20b.mdx @@ -0,0 +1,10 @@ +--- +title: "gpt-oss-20b" +sidebarTitle: "gpt-oss-20b" +description: "Model details for gpt-oss-20b. Summarization and reasoning over supplied context with a 131,072-token context window." +openapi: "api-reference/openapi/playgrounds/gpt-oss-20b.openapi.json POST /chat/completions" +--- + +> OpenAI's gpt-oss-20b is an open-weight Mixture-of-Experts model with 21B total parameters (3.6B active per token), served on ZeroGPU as an MXFP4 build. It is built for summarizing complex information, answering questions from supplied context, and working through multi-step instructions, and it carries a 131,072-token context window. It runs on eligible edge GPUs with cloud fallback, which makes it the lighter and cheaper of ZeroGPU's two gpt-oss models. + +**References:** [Model docs](https://huggingface.co/openai/gpt-oss-20b) • [Terms](https://github.com/openai/gpt-oss/blob/main/LICENSE) • [Privacy](https://zerogpu.ai/privacy-policy) diff --git a/api-reference/models/zlm-v1-tool-decisions.mdx b/api-reference/models/zlm-v1-tool-decisions.mdx index 3e622a2..cce43c7 100644 --- a/api-reference/models/zlm-v1-tool-decisions.mdx +++ b/api-reference/models/zlm-v1-tool-decisions.mdx @@ -5,7 +5,7 @@ description: "Model details for zlm-v1-tool-decisions. MCP tool selection with q openapi: "api-reference/openapi/playgrounds/zlm-v1-tool-decisions.openapi.json POST /responses" --- -> ZeroGPU's tool-decision model turns user intent into action, pairing fast MCP tool selection with query and argument generation. It reads the conversation, selects the tools that fit, and generates the inputs needed to call them, with the decision and the generation happening in a single inference step so an agent moves from understanding a request to acting on it. When no tool is needed it returns `NO_TOOL`, which keeps unnecessary calls out of the workflow. At 0.6B parameters, served as a bf16 build with a 24,576-token context window, it is built for agents that need to choose quickly and act precisely. +> ZeroGPU's tool-decision model turns user intent into action, pairing fast MCP tool selection with query and argument generation. It reads the conversation, selects the tools that fit, and generates the inputs needed to call them, with the decision and the generation happening in a single inference step so an agent moves from understanding a request to acting on it. When no tool is needed it returns `NO_TOOL`, which keeps unnecessary calls out of the workflow. At 0.6B parameters, served as a Q8_0 quantized build with a 24,576-token context window, it is built for agents that need to choose quickly and act precisely. **References:** [Model docs](https://huggingface.co/Qwen/Qwen3-0.6B) • [Terms](https://huggingface.co/Qwen/Qwen3-0.6B/blob/main/LICENSE) • [Privacy](https://zerogpu.ai/privacy-policy) diff --git a/api-reference/openapi/playgrounds/gpt-oss-20b.openapi.json b/api-reference/openapi/playgrounds/gpt-oss-20b.openapi.json new file mode 100644 index 0000000..a60b9ea --- /dev/null +++ b/api-reference/openapi/playgrounds/gpt-oss-20b.openapi.json @@ -0,0 +1,178 @@ +{ + "openapi": "3.1.0", + "info": { + "title": "gpt-oss-20b playground", + "version": "1.0", + "description": "Interactive playground for **gpt-oss-20b**.\nModel is always `gpt-oss-20b` on this page (shown in the form, not editable).\nAuthentication: `x-api-key`, or the OpenAI-compatible `Authorization: Bearer` header. `x-project-id` is optional." + }, + "servers": [ + { + "url": "https://api.zerogpu.ai/v1", + "description": "Production" + } + ], + "security": [ + { + "ApiKey": [] + }, + { + "BearerAuth": [] + } + ], + "paths": { + "/chat/completions": { + "post": { + "operationId": "createChatCompletion_gpt-oss-20b", + "summary": "gpt-oss-20b: Chat completions", + "tags": [ + "gpt-oss-20b" + ], + "requestBody": { + "required": true, + "content": { + "application/json": { + "schema": { + "allOf": [ + { + "$ref": "#/components/schemas/CreateChatCompletionRequest" + }, + { + "type": "object", + "properties": { + "model": { + "type": "string", + "const": "gpt-oss-20b", + "default": "gpt-oss-20b", + "example": "gpt-oss-20b", + "description": "Model identifier (fixed for this playground). Use request examples to change use cases." + } + } + } + ] + }, + "examples": { + "default": { + "summary": "Default", + "value": { + "model": "gpt-oss-20b", + "messages": [ + { + "role": "user", + "content": "The global semiconductor industry is undergoing one of its most significant structural shifts in decades, driven by a combination of geopolitical tensions, surging demand from artificial intelligence workloads, and a wave of government-backed industrial policy across the United States, Europe, and Asia. At the center of this transformation is a race to onshore chip manufacturing capacity that had, for thirty years, been quietly concentrated in Taiwan and South Korea. The United States CHIPS and Science Act, signed into law in 2022, allocated over $52 billion in subsidies to incentivize domestic semiconductor fabrication. Since then, companies including TSMC, Intel, Samsung, and Micron have announced or broken ground on new fabs in Arizona, Ohio, Texas, and Idaho. However, construction timelines have slipped, costs have ballooned, and a shortage of skilled workers has prompted some manufacturers to bring in engineers from overseas — a move that has drawn political scrutiny even as it addresses a genuine talent gap. Meanwhile, the explosion of AI model training and inference has fundamentally altered the demand profile for chips. Graphics processing units originally designed for gaming, particularly those made by NVIDIA, have become the primary compute substrate for large language models. NVIDIA's H100 and successor Blackwell-series GPUs now trade at significant premiums on secondary markets, with some hyperscalers reporting lead times of over a year for large cluster orders. This bottleneck has accelerated investment in custom silicon: Google's Tensor Processing Units, Amazon's Trainium and Inferentia chips, and Meta's MTIA accelerators are all attempts to reduce dependence on third-party GPU supply chains. On the geopolitical front, the United States has progressively tightened export controls on advanced chips and chip-making equipment destined for China, with the most recent rules targeting gate-all-around transistor technology and high-bandwidth memory. China, in response, has dramatically increased state investment in its domestic semiconductor ecosystem, with companies like SMIC and Huawei making incremental but closely watched progress at advanced nodes. Analysts remain divided on how quickly China can close the gap, with estimates ranging from five to fifteen years depending on assumptions about technology transfer, equipment access, and the pace of indigenous innovation. The equipment layer — often overlooked in public discourse — may prove to be the most consequential chokepoint. ASML, the Dutch company that holds a near-monopoly on extreme ultraviolet lithography machines, has become a focal point of both trade diplomacy and industrial strategy. Without EUV tools, fabricating chips below 7 nanometers at scale is effectively impossible, giving ASML and its supply chain extraordinary leverage in a world where nations are suddenly treating semiconductor access as a matter of national security." + } + ] + } + } + } + } + } + }, + "responses": { + "200": { + "description": "Success", + "content": { + "application/json": { + "schema": { + "$ref": "#/components/schemas/ChatCompletionResponse" + } + } + } + }, + "400": { + "description": "Bad request" + }, + "401": { + "description": "Unauthorized" + }, + "403": { + "description": "Forbidden" + }, + "420": { + "description": "Insufficient quota" + }, + "500": { + "description": "Internal server error" + } + }, + "parameters": [ + { + "name": "x-project-id", + "in": "header", + "required": false, + "schema": { + "type": "string" + }, + "description": "Optional project identifier. Scopes the request to a specific project when provided." + } + ] + } + } + }, + "components": { + "securitySchemes": { + "ApiKey": { + "type": "apiKey", + "in": "header", + "name": "x-api-key" + }, + "BearerAuth": { + "type": "http", + "scheme": "bearer", + "description": "OpenAI-compatible: send your ZeroGPU API key as `Authorization: Bearer `." + } + }, + "schemas": { + "CreateChatCompletionRequest": { + "type": "object", + "required": [ + "model", + "messages" + ], + "properties": { + "model": { + "type": "string", + "const": "gpt-oss-20b", + "default": "gpt-oss-20b", + "example": "gpt-oss-20b", + "description": "Model identifier (fixed for this playground). Use request examples to change use cases." + }, + "messages": { + "type": "array", + "minItems": 1, + "items": { + "$ref": "#/components/schemas/ChatMessage" + } + } + } + }, + "ChatMessage": { + "type": "object", + "required": [ + "role", + "content" + ], + "properties": { + "role": { + "type": "string", + "enum": [ + "system", + "user", + "assistant" + ] + }, + "content": { + "type": "string", + "minLength": 1, + "format": "textarea", + "maxLength": 131072, + "description": "Message text." + } + } + }, + "ChatCompletionResponse": { + "type": "object", + "additionalProperties": true + } + } + } +} diff --git a/api-reference/openapi/zerogpu.openapi.json b/api-reference/openapi/zerogpu.openapi.json index f88e640..35f4940 100644 --- a/api-reference/openapi/zerogpu.openapi.json +++ b/api-reference/openapi/zerogpu.openapi.json @@ -1140,6 +1140,7 @@ "LFM2.5-1.2B-Instruct", "LFM2.5-1.2B-Thinking", "gpt-oss-120b", + "gpt-oss-20b", "qwen3-30b-a3b-fp8", "qwen3.6-35b-a3b", "deepseek-v4.1-flash", diff --git a/docs.json b/docs.json index 236a394..4dd4644 100644 --- a/docs.json +++ b/docs.json @@ -104,6 +104,7 @@ "group": "By model", "pages": [ "api-reference/models/gpt-oss-120b", + "api-reference/models/gpt-oss-20b", "api-reference/models/qwen3-30b-a3b-fp8", "api-reference/models/qwen3-6-35b-a3b", "api-reference/models/unlimited-ocr", diff --git a/docs/model-catalog.mdx b/docs/model-catalog.mdx index 7683384..5da47c7 100644 --- a/docs/model-catalog.mdx +++ b/docs/model-catalog.mdx @@ -20,6 +20,7 @@ Sections on this page: [At a glance](#at-a-glance) (pricing table), [Detailed mo | gpt-4.1-mini gpt-4.1-mini | $0.40 | $1.60 | $0.10 | Text Generation | 1,047,576 | | gpt-oss-120b gpt-oss-120b | $0.15 | $0.60 | $0.03 | Text Generation | 131,072 | | qwen3-30b-a3b-fp8 qwen3-30b-a3b-fp8 | $0.10 | $0.45 | $0.05 | Text Generation | 32,768 | +| gpt-oss-20b gpt-oss-20b | $0.02 | $0.10 | $0.01 | Text Generation | 131,072 | | llama-3.1-8b-instruct-fast llama-3.1-8b-instruct-fast | $0.15 | $0.28 | $0.025 | Text Generation | 131,072 | | llama-3.2-3b-instruct-fast llama-3.2-3b-instruct-fast | $0.05 | $0.30 | $0.025 | Text Generation | 131,072 | | llama-guard-4-12b llama-guard-4-12b | $0.18 | $0.18 | — | Text Generation | 163,840 | @@ -33,7 +34,7 @@ Sections on this page: [At a glance](#at-a-glance) (pricing table), [Detailed mo | LFM2.5-1.2B-Thinking LFM2.5-1.2B-Thinking | $0.02 | $0.05 | — | Text Generation | 32,768 | | LFM2.5-1.2B-Instruct LFM2.5-1.2B-Instruct | $0.02 | $0.05 | — | Text Generation | 32,768 | | qwen3.6-35b-a3b qwen3.6-35b-a3b | $0.20 | $1.50 | $0.20 | Text Generation | 16,384 | -| zlm-v1-tool-decisions | $0.05 | $0.15 | — | Text Generation | 24,576 | +| zlm-v1-tool-decisions | $0.04 | Not billed | — | Text Generation | 24,576 | | unlimited-ocr unlimited-ocr | Per page | Per page | — | Text Generation | — | | zlm-v1-signal-extract zlm-v1-signal-extract | $0.04 | $0.10 | — | Text Classification | 400 | | gliner2.5-decide-v1 gliner2.5-decide-v1 | $0.01 | Not billed | — | Text Classification | — | @@ -145,6 +146,16 @@ Sections on this page: [At a glance](#at-a-glance) (pricing table), [Detailed mo Alibaba's Qwen3-30B-A3B is an open-weight Mixture-of-Experts model with 30.5B total parameters (3.3B active per token), served on ZeroGPU as an FP8 build for efficient inference. It thinks through a problem… + +
+ gpt-oss-20b +
gpt-oss-20b
+
+ +
131,072 context window\$0.02 / 1M input\$0.10 / 1M output\$0.01 / 1M cached input
+ + OpenAI's gpt-oss-20b is an open-weight Mixture-of-Experts model with 21B total parameters (3.6B active per token), served on ZeroGPU as an MXFP4 build. It is built for summarizing complex information, answering questions from supplied context… +
llama-3.1-8b-instruct-fast @@ -280,7 +291,7 @@ Sections on this page: [At a glance](#at-a-glance) (pricing table), [Detailed mo
zlm-v1-tool-decisions
-
0.6B params24,576 context window\$0.05 / 1M input\$0.15 / 1M output
+
0.6B params24,576 context window\$0.04 / 1M input
ZeroGPU's tool-decision model turns user intent into action, pairing fast MCP tool selection with query and argument generation. It reads the conversation, selects the tools that fit, and generates the inputs needed to call them…
@@ -370,10 +381,10 @@ Sections on this page: [At a glance](#at-a-glance) (pricing table), [Detailed mo - 10 models: `deepseek-v4.1-flash`, `glm-5.3-flash`, `qwen3-30b-a3b-fp8`, `qwen3.6-35b-a3b`, `gpt-oss-120b`, `llama-3.1-8b-instruct-fast`, `llama-3.2-3b-instruct-fast`, `llama-guard-4-12b`, `zlm-v1-tool-decisions`, `unlimited-ocr` + 11 models: `deepseek-v4.1-flash`, `glm-5.3-flash`, `qwen3-30b-a3b-fp8`, `qwen3.6-35b-a3b`, `gpt-oss-120b`, `gpt-oss-20b`, `llama-3.1-8b-instruct-fast`, `llama-3.2-3b-instruct-fast`, `llama-guard-4-12b`, `zlm-v1-tool-decisions`, `unlimited-ocr` - 16 models: `deepseek-v4.1-flash`, `glm-5.3-flash`, `gpt-5.4-nano`, `gpt-5.6-luna`, `gpt-6-luna`, `gpt-4.1-mini`, `gpt-oss-120b`, `qwen3-30b-a3b-fp8`, `qwen3.6-35b-a3b`, `llama-3.1-8b-instruct-fast`, `llama-3.2-3b-instruct-fast`, `llama-guard-4-12b`, `LFM2.5-1.2B-Thinking`, `LFM2.5-1.2B-Instruct`, `zlm-v1-tool-decisions`, `unlimited-ocr` + 17 models: `deepseek-v4.1-flash`, `glm-5.3-flash`, `gpt-5.4-nano`, `gpt-5.6-luna`, `gpt-6-luna`, `gpt-4.1-mini`, `gpt-oss-120b`, `gpt-oss-20b`, `qwen3-30b-a3b-fp8`, `qwen3.6-35b-a3b`, `llama-3.1-8b-instruct-fast`, `llama-3.2-3b-instruct-fast`, `llama-guard-4-12b`, `LFM2.5-1.2B-Thinking`, `LFM2.5-1.2B-Instruct`, `zlm-v1-tool-decisions`, `unlimited-ocr` 3 models: `whisper-tiny`, `chatterbox-nano`, `chatterbox` diff --git a/docs/open-weight.mdx b/docs/open-weight.mdx index afd7531..6f67336 100644 --- a/docs/open-weight.mdx +++ b/docs/open-weight.mdx @@ -3,7 +3,7 @@ title: "Open Weight" description: "The open-weight frontier-class models served on ZeroGPU, with pricing and context windows." --- -Open-weight models ship with published weights and permissive licenses, so you can inspect them, self-host them, and avoid lock-in to a closed vendor. ZeroGPU serves ten of them — for reasoning, coding, agentic workflows, multimodal understanding, high-volume summarization, document OCR, tool selection, and safety classification — through the same OpenAI-compatible API as every other model in the [catalog](/docs/model-catalog). +Open-weight models ship with published weights and permissive licenses, so you can inspect them, self-host them, and avoid lock-in to a closed vendor. ZeroGPU serves eleven of them — for reasoning, coding, agentic workflows, multimodal understanding, high-volume summarization, document OCR, tool selection, and safety classification — through the same OpenAI-compatible API as every other model in the [catalog](/docs/model-catalog). Full request and response examples live on the [Text Generation](/docs/text-generation) page. Open-weight speech-to-text and text-to-speech models are on [Audio Models](/docs/audio). @@ -16,10 +16,11 @@ Full request and response examples live on the [Text Generation](/docs/text-gene | qwen3-30b-a3b-fp8 qwen3-30b-a3b-fp8 | $0.10 | $0.45 | $0.05 | 32,768 | | qwen3.6-35b-a3b qwen3.6-35b-a3b | $0.20 | $1.50 | $0.20 | 16,384 | | gpt-oss-120b gpt-oss-120b | $0.15 | $0.60 | $0.03 | 131,072 | +| gpt-oss-20b gpt-oss-20b | $0.02 | $0.10 | $0.01 | 131,072 | | llama-3.1-8b-instruct-fast llama-3.1-8b-instruct-fast | $0.15 | $0.28 | $0.025 | 131,072 | | llama-3.2-3b-instruct-fast llama-3.2-3b-instruct-fast | $0.05 | $0.30 | $0.025 | 131,072 | | llama-guard-4-12b llama-guard-4-12b | $0.18 | $0.18 | — | 163,840 | -| zlm-v1-tool-decisions | $0.05 | $0.15 | — | 24,576 | +| zlm-v1-tool-decisions | $0.04 | Not billed | — | 24,576 | | unlimited-ocr unlimited-ocr | Per page | Per page | — | — | ## Model cards @@ -75,6 +76,16 @@ Full request and response examples live on the [Text Generation](/docs/text-gene OpenAI's gpt-oss-120b is an open-weight Mixture-of-Experts model with 117B total parameters (5.1B active per token), served on ZeroGPU for general text generation. It reasons through a problem… + +
+ gpt-oss-20b +
gpt-oss-20b
+
+ +
131,072 context window\$0.02 / 1M input\$0.10 / 1M output\$0.01 / 1M cached input
+ + OpenAI's gpt-oss-20b is an open-weight Mixture-of-Experts model with 21B total parameters (3.6B active per token), served on ZeroGPU as an MXFP4 build. It is built for summarizing complex information, answering questions from supplied context… +
llama-3.1-8b-instruct-fast @@ -110,7 +121,7 @@ Full request and response examples live on the [Text Generation](/docs/text-gene
zlm-v1-tool-decisions
-
0.6B params24,576 context window\$0.05 / 1M input\$0.15 / 1M output
+
0.6B params24,576 context window\$0.04 / 1M input
ZeroGPU's tool-decision model turns user intent into action, pairing fast MCP tool selection with query and argument generation. It reads the conversation, selects the tools that fit, and generates the inputs needed to call them…
diff --git a/docs/text-generation.mdx b/docs/text-generation.mdx index f370237..ed420b6 100644 --- a/docs/text-generation.mdx +++ b/docs/text-generation.mdx @@ -3,7 +3,7 @@ title: "Text Generation" description: "Edge generative models for conversation and reasoning." --- -Text generation models produce new text from an input prompt. ZeroGPU offers open-weight heavyweights — DeepSeek's deepseek-v4.1-flash, OpenAI's gpt-oss-120b, Alibaba's qwen3-30b-a3b-fp8 and multimodal qwen3.6-35b-a3b, and Z.ai's glm-5.3-flash — for reasoning and tool use, OpenAI's gpt-5.4-nano, gpt-5.6-luna, gpt-6-luna, and gpt-4.1-mini for cost-efficient high-volume work, plus conversational and reasoning models from Liquid AI, Meta's llama-3.1-8b-instruct-fast and llama-3.2-3b-instruct-fast for high-volume summarization, Meta's llama-guard-4-12b for safety classification, ZeroGPU's zlm-v1-tool-decisions for picking an agent's tools and generating their arguments, and Baidu's unlimited-ocr for turning PDFs and scanned documents into Markdown. Each request is shown for both the [Responses API](/api-reference/responses) and the OpenAI-compatible [Chat Completions API](/api-reference/chat-completions) — except `qwen3-30b-a3b-fp8` and `llama-3.2-3b-instruct-fast`, which are Chat Completions only. +Text generation models produce new text from an input prompt. ZeroGPU offers open-weight heavyweights — DeepSeek's deepseek-v4.1-flash, OpenAI's gpt-oss-120b and gpt-oss-20b, Alibaba's qwen3-30b-a3b-fp8 and multimodal qwen3.6-35b-a3b, and Z.ai's glm-5.3-flash — for reasoning and tool use, OpenAI's gpt-5.4-nano, gpt-5.6-luna, gpt-6-luna, and gpt-4.1-mini for cost-efficient high-volume work, plus conversational and reasoning models from Liquid AI, Meta's llama-3.1-8b-instruct-fast and llama-3.2-3b-instruct-fast for high-volume summarization, Meta's llama-guard-4-12b for safety classification, ZeroGPU's zlm-v1-tool-decisions for picking an agent's tools and generating their arguments, and Baidu's unlimited-ocr for turning PDFs and scanned documents into Markdown. Each request is shown for both the [Responses API](/api-reference/responses) and the OpenAI-compatible [Chat Completions API](/api-reference/chat-completions) — except `qwen3-30b-a3b-fp8`, `llama-3.2-3b-instruct-fast`, and `gpt-oss-20b`, which are Chat Completions only. | Model | Input /1M | Output /1M | Cached input /1M | Max tokens | | --- | ---: | ---: | ---: | ---: | @@ -15,13 +15,14 @@ Text generation models produce new text from an input prompt. ZeroGPU offers ope | gpt-5.6-luna gpt-5.6-luna | $0.20 | $1.20 | $0.20 | 272,000 | | gpt-6-luna gpt-6-luna | $0.10 | $0.50 | $0.01 | 272,000 | | gpt-4.1-mini gpt-4.1-mini | $0.40 | $1.60 | $0.10 | 1,047,576 | +| gpt-oss-20b gpt-oss-20b | $0.02 | $0.10 | $0.01 | 131,072 | | llama-3.1-8b-instruct-fast llama-3.1-8b-instruct-fast | $0.15 | $0.28 | $0.025 | 131,072 | | llama-3.2-3b-instruct-fast llama-3.2-3b-instruct-fast | $0.05 | $0.30 | $0.025 | 131,072 | | llama-guard-4-12b llama-guard-4-12b | $0.18 | $0.18 | — | 163,840 | | LFM2.5-1.2B-Instruct LFM2.5-1.2B-Instruct | $0.02 | $0.05 | — | 32,768 | | LFM2.5-1.2B-Thinking LFM2.5-1.2B-Thinking | $0.02 | $0.05 | — | 32,768 | | qwen3.6-35b-a3b qwen3.6-35b-a3b | $0.20 | $1.50 | $0.20 | 16,384 | -| zlm-v1-tool-decisions | $0.05 | $0.15 | — | 24,576 | +| zlm-v1-tool-decisions | $0.04 | Not billed | — | 24,576 | | unlimited-ocr unlimited-ocr | Per page | Per page | — | — | ## gpt-oss-120b @@ -418,6 +419,34 @@ curl https://api.zerogpu.ai/v1/chat/completions \ The model returns its answer as the assistant message — an `output_text` item in Responses, the `message.content` field in Chat Completions — with `usage` token counts alongside it. +## gpt-oss-20b + + + `gpt-oss-20b` is routable on `POST /v1/chat/completions` only. It is not currently + available on `/v1/responses`. + + +> OpenAI's gpt-oss-20b is an open-weight Mixture-of-Experts model with 21B total parameters (3.6B active per token), served on ZeroGPU as an MXFP4 build. It is built for summarizing complex information, answering questions from supplied context, and working through multi-step instructions, and it carries a 131,072-token context window. It runs on eligible edge GPUs with cloud fallback, which makes it the lighter and cheaper of ZeroGPU's two gpt-oss models. + +**References:** [Model docs](https://huggingface.co/openai/gpt-oss-20b) • [Terms](https://github.com/openai/gpt-oss/blob/main/LICENSE) • [Privacy](https://zerogpu.ai/privacy-policy) + +```bash Chat Completions +curl https://api.zerogpu.ai/v1/chat/completions \ + -H "content-type: application/json" \ + -H "x-api-key: YOUR_API_KEY" \ + -d '{ + "model": "gpt-oss-20b", + "messages": [ + { + "role": "user", + "content": "The global semiconductor industry is undergoing one of its most significant structural shifts in decades, driven by a combination of geopolitical tensions, surging demand from artificial intelligence workloads, and a wave of government-backed industrial policy across the United States, Europe, and Asia. At the center of this transformation is a race to onshore chip manufacturing capacity that had, for thirty years, been quietly concentrated in Taiwan and South Korea. The United States CHIPS and Science Act, signed into law in 2022, allocated over $52 billion in subsidies to incentivize domestic semiconductor fabrication. Since then, companies including TSMC, Intel, Samsung, and Micron have announced or broken ground on new fabs in Arizona, Ohio, Texas, and Idaho. However, construction timelines have slipped, costs have ballooned, and a shortage of skilled workers has prompted some manufacturers to bring in engineers from overseas — a move that has drawn political scrutiny even as it addresses a genuine talent gap. Meanwhile, the explosion of AI model training and inference has fundamentally altered the demand profile for chips. Graphics processing units originally designed for gaming, particularly those made by NVIDIA, have become the primary compute substrate for large language models. NVIDIA's H100 and successor Blackwell-series GPUs now trade at significant premiums on secondary markets, with some hyperscalers reporting lead times of over a year for large cluster orders. This bottleneck has accelerated investment in custom silicon: Google's Tensor Processing Units, Amazon's Trainium and Inferentia chips, and Meta's MTIA accelerators are all attempts to reduce dependence on third-party GPU supply chains. On the geopolitical front, the United States has progressively tightened export controls on advanced chips and chip-making equipment destined for China, with the most recent rules targeting gate-all-around transistor technology and high-bandwidth memory. China, in response, has dramatically increased state investment in its domestic semiconductor ecosystem, with companies like SMIC and Huawei making incremental but closely watched progress at advanced nodes. Analysts remain divided on how quickly China can close the gap, with estimates ranging from five to fifteen years depending on assumptions about technology transfer, equipment access, and the pace of indigenous innovation. The equipment layer — often overlooked in public discourse — may prove to be the most consequential chokepoint. ASML, the Dutch company that holds a near-monopoly on extreme ultraviolet lithography machines, has become a focal point of both trade diplomacy and industrial strategy. Without EUV tools, fabricating chips below 7 nanometers at scale is effectively impossible, giving ASML and its supply chain extraordinary leverage in a world where nations are suddenly treating semiconductor access as a matter of national security." + } + ] +}' +``` + +The summary is returned as `choices[].message.content`. + ## llama-3.1-8b-instruct-fast > Meta's Llama 3.1 Instruct, tuned for fast, low-cost summarization at scale on the ZeroGPU edge network. Its 128K-token context window takes in entire documents, long transcripts, and full email or chat threads in a single pass. No chunking, no stitching partial summaries back together. Built for the high-volume summarization that runs around your core product: condensing long reports and transcripts, summarizing email and chat threads, and processing articles in bulk. When context length matters and you're summarizing on every request, this is the model that keeps it fast and affordable. @@ -706,7 +735,7 @@ The model returns a text answer describing the images and video it was given. ## zlm-v1-tool-decisions -> ZeroGPU's tool-decision model turns user intent into action, pairing fast MCP tool selection with query and argument generation. It reads the conversation, selects the tools that fit, and generates the inputs needed to call them, with the decision and the generation happening in a single inference step so an agent moves from understanding a request to acting on it. When no tool is needed it returns `NO_TOOL`, which keeps unnecessary calls out of the workflow. At 0.6B parameters, served as a bf16 build with a 24,576-token context window, it is built for agents that need to choose quickly and act precisely. +> ZeroGPU's tool-decision model turns user intent into action, pairing fast MCP tool selection with query and argument generation. It reads the conversation, selects the tools that fit, and generates the inputs needed to call them, with the decision and the generation happening in a single inference step so an agent moves from understanding a request to acting on it. When no tool is needed it returns `NO_TOOL`, which keeps unnecessary calls out of the workflow. At 0.6B parameters, served as a Q8_0 quantized build with a 24,576-token context window, it is built for agents that need to choose quickly and act precisely. **References:** [Model docs](https://huggingface.co/Qwen/Qwen3-0.6B) • [Terms](https://huggingface.co/Qwen/Qwen3-0.6B/blob/main/LICENSE) • [Privacy](https://zerogpu.ai/privacy-policy)