gpt-oss-120B API: Pricing, Benchmarks & Docs | Together AI (https://www.together.ai/models/gpt-oss-120b) citeturn151view2 [wordlim: 200] Crawled: today; Content type: text/html; Source: find({"ref_id":"turn150view1","pattern":"$0.15"}); Total lines: 339 L0: 【0†iframe†www.googletagmanager.com】 L1: L2: * 【1†Inference】 L3: L4: * * * * * 【2†Dedicated Container Inference Inference for custom models】 L5: * 【3†New Polymorphic Inference Inference that gets better over time】 L6: L7: Explore Top Open-Source Models L8: L9: * * * * * 【4†Show all】 L10: L11: * 【1†Compute】 L12: L13: * 【1†Model Shaping】 L14: L15: * 【1†Research】 L16: L17: * 【1†Developers】 L18: L19: * 【5†Customers】 L20: * 【6†Pricing】 L21: * * * * * * 【2†Dedicated Container Inference Inference for custom models】 L22: * 【3†New Polymorphic Inference Inference that gets better over time】 L23: L24: Explore Top Open-Source Models L25: L26: * * * * * 【4†Show all】 L27: L28: * * * * 【7†Contact sales】 L29: L30: 【7†Contact sales】 L31: L32: 【8†Sign in†api.together.ai】 L33: L34: 【4†Models】 / OpenAI L35: L36: Reasoning L37: L38: Chat L39: # gpt-oss-120B L40: L41: Advanced open reasoning model with enterprise-grade capabilities. L42: L43: 【1†Try now】【9†Read Docs†docs.together.ai】 L44: L45: 【10†Image†cdn.prod.website-files.com】 L46: ## About model L47: L48: gpt-oss-120B has 120B parameters and a 128K-token context window. On Together AI it costs $0.15 per 1M input tokens and $0.60 per 1M output tokens. L49: Enterprise-Ready Open Reasoning: L50: gpt-oss-120B delivers sophisticated chain-of-thought reasoning capabilities in a fully open model. Built with community feedback and released under Apache 2.0, this 120B parameter model provides transparency, customization, and deployment flexibility for organizations requiring complete data security & privacy control. L51: L52: Quickstart guides L53: L54: 【11†RAG Building a RAG Workflow†docs.together.ai】 L55: L56: 【12†Agents Agent Workflows†docs.together.ai】 L57: 【13†Apps Next.js Chat Quickstart†docs.together.ai】 L58: L59: Performance benchmarks L60: Model | FrontierMath Tier 4 | GPQA Diamond | HLE | SciCode | GDPval-AA | Terminal-Bench 2.1 | Agent Arena | FrontierCode | DeepSWE L61: --- | --- | --- | --- | --- | --- | --- | --- | --- | --- L62: 【10†Image†cdn.prod.website-files.com】gpt-oss-120B | | 78.2% | 18% | 39% | 15% | 26% | | | L63: Related open-source models L64: --- L65: Competitor closed-source models L66: --- L67: 【14†Image: Claude Fable 5†cdn.prod.website-files.com】Claude Fable 5 | 87.8% | 92.6% | 53% | 60% | 62% | 85% | | 53.5% | 70% L68: 【14†Image: Claude Opus 5†cdn.prod.website-files.com】Claude Opus 5 | 73.2% | 93.2% | 53% | 56% | 68% | 89% | +16.4pp | 53.4% | 74% L69: 【10†Image: GPT-5.6 Sol†cdn.prod.website-files.com】GPT-5.6 Sol | 82.9% | 94.1% | 47% | 56% | 61% | 88% | | 47.5% | 73% L70: 【15†Image: Grok 4.5†cdn.prod.website-files.com】Grok 4.5 | 24.4% | 93.1% | 40% | 54% | 51% | 82% | +4.2pp | 42.4% | 54% L71: 【10†Image: GPT-5.6 Luna†cdn.prod.website-files.com】GPT-5.6 Luna | 61.0% | 91.1% | 37% | 53% | 54% | 81% | | 39.8% | 67% L72: * ### API usage L73: L74: * cURL L75: L76: * Python L77: L78: * Typescript L79: L80: Endpoint: L81: openai/gpt-oss-120b L82: L83: `curl -X POST "https://api.together.ai/v1/chat/completions" \ L84: -H "Authorization: Bearer $TOGETHER_API_KEY" \ L85: -H "Content-Type: application/json" \ L86: -d '{ L87: "model": "openai/gpt-oss-120b", L88: "messages": [ L89: { L90: "role": "user", L91: "content": "What are some fun things to do in New York?" L92: } L93: ] L94: }'` L95: L96: L97: `from together import Together L98: L99: client = Together() L100: response = client.chat.completions.create( L101: model="openai/gpt-oss-120b", L102: messages=[ L103: { L104: "role": "user", L105: "content": "What are some fun things to do in New York?" L106: } L107: ] L108: ) L109: print(response.choices[0].message.content)` L110: L111: L112: `import Together from 'together-ai'; L113: const together = new Together(); L114: L115: const completion = await together.chat.completions.create({ L116: model: 'openai/gpt-oss-120b', L117: messages: [ L118: { L119: role: 'user', L120: content: 'What are some fun things to do in New York?' L121: } L122: ], L123: }); L124: L125: console.log(completion.choices[0].message.content);` L126: * ### Model card L127: Architecture Overview: L128: • Mixture-of-Experts (MoE) architecture with SwiGLU activations L129: • Alternating attention layers between full context and sliding 128-token window L130: • Learned attention sink per-head for enhanced performance L131: Training Methodology: L132: • Comprehensive safety training and evaluation protocols L133: • Community feedback integration from global listening sessions L134: • Rigorous testing under Preparedness Framework L135: • Standard GPT-4o tokenizer with additional Harmony format tokens L136: Performance Characteristics: L137: • Native FP4 quantization for efficient inference L138: • 128K context window with RoPE positional encoding L139: • Chain-of-thought reasoning with adjustable effort levels L140: * ### Applications & use cases L141: Enterprise Applications: L142: • Complex reasoning and analysis tasks L143: • Research and development support L144: • Technical documentation generation L145: • Strategic planning and decision support L146: Developer Use Cases: L147: • Code generation and review L148: • API development and integration L149: • System architecture design L150: • Technical troubleshooting and debugging L151: Industry Solutions: L152: • Healthcare: Clinical decision support and medical research L153: • Finance: Risk analysis and regulatory compliance L154: • Legal: Contract analysis and legal research L155: • Education: Curriculum development and tutoring systems L156: Deployment Scenarios: L157: • On-premises infrastructure for data sovereignty L158: • Private cloud deployments for security compliance L159: • Custom fine-tuning for domain-specific applications L160: • Multi-modal integration with existing systems L161: 【1†Model specifications】【1†Model data】 L162: L163: * Model provider L164: L165: 【10†Image†cdn.prod.website-files.com】 L166: L167: OpenAI L168: L169: * Type L170: L171: Reasoning L172: L173: Chat L174: L175: * Main use cases L176: L177: Chat L178: L179: Small & Fast L180: L181: Medium General Purpose L182: L183: * Features L184: L185: JSON Mode L186: L187: * Speed L188: L189: High L190: L191: * Intelligence L192: L193: High L194: L195: * Deployment L196: L197: 【16†Serverless】 L198: L199: 【17†Dedicated】 L200: L201: * Endpoint L202: L203: 【1†openai/gpt-oss-120b】 L204: L205: * Parameters L206: L207: 120B L208: L209: * Context length L210: L211: 128K L212: L213: * Input price L214: L215: $0.15 / 1M tokens L216: L217: * Output price L218: L219: $0.60 / 1M tokens L220: L221: * Input modalities L222: L223: Text L224: * Output modalities L225: L226: Text L227: L228: * Released L229: L230: August 4, 2025 L231: L232: * Last updated L233: L234: August 18, 2025 L235: L236: * External link L237: L238: 【18†Provider docs†huggingface.co】 L239: L240: * Category L241: L242: Chat L243: L244: 【19†Run in Playground†api.together.ai】 L245: L246: 【9†Quickstart docs†docs.together.ai】【1†Deploy Model】 L247: ## Start building on Together AI L248: L249: From optimized training and model shaping to large-scale production inference L250: L251: 【4†Explore models】 L252: L253: 【20†Image†cdn.prod.website-files.com】 L254: L255: * Products L256: L257: * 【16†Serverless Inference】 L258: * 【21†Batch Inference】 L259: * 【22†Provisioned Throughput】 L260: * 【17†Dedicated Model Inference】 L261: * 【2†Dedicated Container Inference】 L262: * 【23†Together Link】 L263: * 【24†Fine-Tuning】 L264: * 【25†Post-Training】 L265: * 【26†Evaluations】 L266: * 【27†GPU Clusters】 L267: * 【28†Sandbox】 L268: L269: * Models L270: 【4†See all models】 L271: L272: 【29†ZAI】 L273: L274: 【30†Moonshot AI】 L275: L276: 【31†MiniMax AI】 L277: L278: 【32†DeepSeek】 L279: L280: 【33†Qwen】 L281: L282: 【34†Meta】 L283: L284: 【35†Google】 L285: L286: 【36†OpenAI】 L287: L288: 【37†Mistral AI】 L289: L290: 【2†Custom models】 L291: L292: * Developers L293: L294: * 【38†Research】 L295: * 【39†Docs†docs.together.ai】 L296: * 【40†Open-source AI】 L297: * 【41†OSS ROI calculator】 L298: * 【42†Voice Agents】 L299: L300: Pricing L301: L302: * 【6†Pricing overview】 L303: * 【43†Inference】 L304: * 【44†Fine-Tuning】 L305: * 【45†GPU Clusters】 L306: L307: * Resources L308: * 【46†Blog】 L309: * 【5†Customer Stories】 L310: * 【47†Startup Accelerator】 L311: * 【48†Support】 L312: L313: Company L314: L315: * 【49†About us】 L316: * 【50†Careers】 L317: * 【51†Events】 L318: * 【52†Press】 L319: L320: * 【53†Privacy Policy】 L321: * 【54†Terms of service】 L322: * 【55†Cookie Policy】 L323: * 【1†Consent Preferences】 L324: L325: © 2026 Together AI. All Rights Reserved. L326: L327: * * * 【56†Image†cdn.prod.website-files.com】 L328: L329: Launching soon L330: L331: Want to be notified when the model is available on Together AI? L332: L333: You're on the list L334: We'll email you when the endpoint goes live. Until then, 200+ models are ready to call. L335: L336: 【4†Browse Models】【1†Continue】 L337: L338: 【57†Image†px.ads.linkedin.com】