GPT-OSS 120B vs Kimi K3
Benchmarks, API Token Pricing, Context Window & Speed Comparison
Related topics (4)
Compare GPT-OSS 120B by OpenAI against Kimi K3 by Moonshot AI across real-world coding, reasoning, token economics, and latency.
切换并对比任意模型
Select any AI models from our verified benchmark database to generate live side-by-side matrices.
Live comparison: no in-depth editorial review has been published for this pair yet.
GPT-OSS 120B
开源权重基础大模型OpenAI 推出的 GPT-OSS 120B 是一款支持私有云服务器自托管、微调与完全数据主权的开源权重基础大模型，具备 131K 上下文窗口、95 Token/秒生成速度与 56/100 综合智能指数（每百万 Token $0.15 输入 / $0.6 输出）。
$0.15 / 1M
$0.60 / 1M
131.1K tokens
45.5%
Kimi K3
开源权重推理大模型Moonshot AI 推出的 Kimi K3 是一款通过大规模强化学习训练、支持私有化部署的开源权重推理大模型，具备 1.0M 上下文窗口、58 Token/秒生成速度与 93/100 综合智能指数（免费开源权重）。
$0.00 / 1M
$0.00 / 1M
1.05M tokens
74.5%
核心差异一览
- Kimi K3 is 1.0x cheaper for API usage (input + output tokens combined).
- Kimi K3 has a larger context window (1M vs 131K tokens).
- GPT-OSS 120B is faster at 95 tok/s vs 58 tok/s.
- Kimi K3 leads on SWE-bench coding (74.5% vs 45.5%).
| Feature | GPT-OSS 120B | Kimi K3 |
|---|---|---|
Intelligence Index Composite reasoning & knowledge score (0-100) | 56 | 93 |
Provider | OpenAI | Moonshot AI |
Input Price Cost per million input tokens | $0.15 / 1M | $0.00 / 1M |
Output Price Cost per million output tokens | $0.60 / 1M | $0.00 / 1M |
Blended Price Weighted 7:2:1 cache:input:output per 1M | $0.10 / 1M | $0.00 / 1M |
Context Window Maximum input + output tokens | 131.1K tokens~197 A4 pages of text | 1.05M tokens~1,573 A4 pages of text |
Max Output Tokens Per single request | 33K | 131K |
Throughput Speed Tokens generated per second | 95 tok/s | 58 tok/s |
SWE-bench Verified Real-world software engineering | 45.5% | 74.5% |
GPQA Diamond Graduate-level reasoning | 68.5% | 79% |
Cached Input Price Prompt caching discount per 1M tokens | $0.01 / 1M | $0.00 / 1M |
各项指标谁更胜一筹？
At-a-glance decision framework权威基准测试得分对比
Official evaluation metrics verified by independent third-party research labs.
Industry benchmark evaluating real-world GitHub issue resolution and multi-file debugging.
Challenging questions created by domain experts to test complex PhD-level reasoning.
Hardened reasoning and academic knowledge benchmark with reduced lucky-guess probability.
Crowdsourced human preference evaluation based on blind, head-to-head model battles.
Extremely difficult, closed-ended questions spanning the frontier of human expert knowledge.
Autonomous multi-step tasks executed inside a real command-line terminal environment.
Research-grade coding problems drawn from real scientific and numerical workloads.
Contamination-free competitive programming problems collected continuously over time.
American Invitational Mathematics Examination — multi-step Olympiad-level reasoning.
Accuracy when reasoning over very large inputs that fill the model context window.
Tokens emitted per second during active generation streaming.
延迟与响应速度
Time to first token (TTFT) and estimated end-to-end response time for a typical 500-token response.
实时 Token 成本模拟器
Adjust sliders to calculate projected monthly cloud API spending with prompt caching discounts.
GPT-OSS 120B · Strengths & Key Advantages
- 提供 131K 上下文窗口与最高 33K 输出容量，核心基准表现优异（SWE-bench Verified 45.5% · GPQA Diamond 68.5%）
- 高达 95 Token/秒的生成吞吐量，首字延迟 (TTFT) 仅约 480 毫秒
- 开放模型权重 (117B (5.1B MoE))，支持通过 vLLM、SGLang 或 Ollama 在私有 GPU 集群上本地部署与微调
Considerations & Trade-offs
- 在本地以满血精度运行需要较高的 GPU 显存 (VRAM) 与运维基础设施
- 在超长链路的前沿科研推导任务上，深度略逊于顶配重型旗舰推理模型
Kimi K3 · Strengths & Key Advantages
- 提供 1.0M 上下文窗口与最高 131K 输出容量，核心基准表现优异（SWE-bench Verified 74.5% · GPQA Diamond 79%）
- 高达 58 Token/秒的生成吞吐量，首字延迟 (TTFT) 仅约 720 毫秒
- 开放模型权重 (2.8T total (MoE))，支持通过 vLLM、SGLang 或 Ollama 在私有 GPU 集群上本地部署与微调
Considerations & Trade-offs
- 在本地以满血精度运行需要较高的 GPU 显存 (VRAM) 与运维基础设施
- 在超长链路的前沿科研推导任务上，深度略逊于顶配重型旗舰推理模型
Token Use & Task Economics
Estimated cost and token consumption to complete one representative agentic task.
| Metric | GPT-OSS 120B | Kimi K3 |
|---|---|---|
Cost per Task Estimated USD to finish one task | $0.14 | $2.00 |
Output Tokens / Task Tokens generated per task | 45K | 35K |
Reasoning Tokens / Task Thinking tokens burned per task | 22K | 18K |
技术规格参数
Architecture, release timeline, capabilities, and modalities.
| Feature | GPT-OSS 120B | Kimi K3 |
|---|---|---|
| Official API Model ID (SDK) | gpt-oss-120b | kimi-k3 |
| API Protocol | OpenAI Responses API & Chat Completions v1 | OpenAI-Compatible REST / vLLM / SGLang / HuggingFace TGI |
| Release Date | 2025-08 | 2026-07 |
| Knowledge Cutoff | June 2024 | March 2026 |
| Architecture | Sparse MoE transformer with reasoning | Colossus-Scale Multimodal Sparse Mixture-of-Experts |
| Parameters | 117B (5.1B active MoE) | 2.8T total (MoE) |
| Enterprise Rate Limits (RPM / TPM) | Unlimited (Hardware-bound on self-hosted GPU) / 10,000 RPM Cloud · Unlimited local throughput / 10M+ TPM Managed Cloud | Unlimited (Hardware-bound on self-hosted GPU) / 10,000 RPM Cloud · Unlimited local throughput / 10M+ TPM Managed Cloud |
| Long-Context Tier (>128K/200K) | Flat rate ($0.15/1M in) across full 131K window | Free across full context window (only KV-cache VRAM scales) |
| Prompt Cache Policy & TTL | Automatic prefix caching at $0.015/1M (no extra cache-write fee) | Automatic prefix KV-caching in vLLM / SGLang (0 cost) |
| Computer Use / GUI Agent | API / CLI Tool-Use & DOM Agent ready (no raw pixel desktop click loop) | Supported (Screen coordinate grounding, browser & desktop GUI action loops) |
| MCP & Parallel Tool Calling | Native MCP Client/Server ready · Parallel function calling · 100% Strict JSON Schema | Native MCP Client/Server ready · Parallel function calling · 100% Strict JSON Schema |
| Web Search Grounding & Sandbox | Native Web Search tool + Hosted Code Interpreter & File Search RAG | Supported via external Tavily/Exa Search & E2B/Docker Sandbox tools |
| Official Fine-Tuning Support | Full Weights SFT, LoRA, QLoRA, DPO & GRPO RL (Unsloth / Axolotl / Llama-Factory) | Full Weights SFT, LoRA, QLoRA, DPO & GRPO RL (Unsloth / Axolotl / Llama-Factory) |
| Multimodal Token Billing | Text BPE (~4 chars/token) | Text BPE (~4 chars/token) · Vision (~258–765 tokens per image tile) |
| Quantization & Min GPU VRAM | BF16 / FP16, FP8 Dynamic, INT4 AWQ / GPTQ, GGUF (Q4_K_M / Q8_0) · 73 GB VRAM (4-bit Quantized) · 252 GB VRAM (Native BF16) | BF16 / FP16, FP8 Dynamic, INT4 AWQ / GPTQ, GGUF (Q4_K_M / Q8_0) · 1736 GB VRAM (4-bit Quantized) · 6020 GB VRAM (Native BF16) |
| Safety & Frontier Classification | OpenAI Preparedness Framework Audited & System Card Published | Open-Weights System Card + Customizable Guardrail Layer (Llama Guard / ShieldGemma) |
| Enterprise Compliance (ZDR/HIPAA/SOC2) | 100% Air-Gapped On-Premises / Private VPC · Full HIPAA, GDPR & SOC2 Control | 100% Air-Gapped On-Premises / Private VPC · Full HIPAA, GDPR & SOC2 Control |
| Input Modalities | text | text, vision |
| Output Modalities | text | text |
GPT-OSS 120B vs Kimi K3: 常见问题解答 (FAQ)
Which is more intelligent, GPT-OSS 120B or Kimi K3?
Kimi K3 is more intelligent. It scores 93 compared with 56 on the composite Intelligence Index.
Which model is faster?
GPT-OSS 120B is faster, generating 95 tokens per second compared with 58 tokens per second.
Which model is cheaper to run?
Kimi K3 is cheaper at $0.00 per 1M tokens blended, compared with $0.10 (7:2:1 cache:input:output ratio).
Which model has lower latency?
GPT-OSS 120B has lower latency, with a time to first token of 0.48s compared with 0.72s.
Which model has a larger context window?
Kimi K3 has a larger context window, supporting 1.05M tokens compared with 131.1K tokens.
综合结论
Choose GPT-OSS 120B for 支持私有云服务器自托管、微调与完全数据主权的开源权重基础任务、企业级 rag 检索增强生成、自动化智能体工作流与高可靠 api 集成。, or choose Kimi K3 for 通过大规模强化学习训练、支持私有化部署的开源权重推理任务、企业级 rag 检索增强生成、自动化智能体工作流与高可靠 api 集成。.
常见问题 (FAQ)
Which is cheaper: GPT-OSS 120B or Kimi K3?+
GPT-OSS 120B costs $0.15/1M input and $0.6/1M output tokens, while Kimi K3 costs $0/1M input and $0/1M output tokens.
Which model has a larger context window: GPT-OSS 120B or Kimi K3?+
GPT-OSS 120B supports 131K tokens whereas Kimi K3 supports 1.0M tokens.
Understand key metrics like context window, token pricing, reasoning mode, and SWE-bench scores in our comprehensive AI glossary.
Related reading
- Aion-3.0 vs Aion-Mini：基准测试、API 价格与速度全面对比 (2026)对比
- Aion-3.0 vs Gemini 3.6 Flash：基准测试、API 价格与速度全面对比 (2026)对比
- Aion-Mini vs Laguna XS 2.1：基准测试、API 价格与速度全面对比 (2026)对比
- Aya Expanse 32B vs Llama 3.2 90B Vision：基准测试、API 价格与速度全面对比 (2026)对比
- Aya Expanse 8B vs Command R+：基准测试、API 价格与速度全面对比 (2026)对比
- Aya Expanse 8B vs Qwen 2.5 7B Instruct：基准测试、API 价格与速度全面对比 (2026)对比
Benchmark scores, token prices, and rate limits evolve rapidly as AI labs release model updates. Always verify current specifications and pricing on the official provider documentation.
数据来源: OpenAI & Moonshot AI Official Documentation