DeepSeek R1
Core Facts
GPT-4和Mixtral等主流大模型也采用了MoE架构
DeepSeek R1 does not support Function Calling by default, whereas DeepSeek V3 supports Function Calling.
DeepSeek R1 offers multiple distilled versions ranging from 1.5B to 671B parameters.
Timeline (last 90 days)
The release of DeepSeek R1 forced the industry to rethink the price floor for open-source models.
Open-source models scored lower: Qwen3 235B scored 11/12, both sizes of Gemma 4 scored 9/12, DeepSeek R1 scored 7/12, and GLM-5 scored 5/12.
The models tested include Google's Gemini 3.7 Flash, Z.ai's GLM-5, Alibaba's Qwen 3 235B A22B Instruct, and OpenAI's gpt-oss-120b, all running on Kaggle.
The inexpensive Gemini 3.7 Flash and the open-source DeepSeek R1 both outperformed the two Claude flagship models in following director's notes.
Two months after the previous-generation 27B model's release, it became the most liked model in Hugging Face history, surpassing DeepSeek R1 and Llama in popularity.
The so-called locally deployed DeepSeek actually runs a distilled small model (the R1 distilled version).
After the author connected the Exa web search MCP server to TrueForge, the agent produced a research summary via real web searches that included models such as DeepSeek R1, MiniMax, GLM, Mistral, and others.
OpenAI used Semi-Analysis's public benchmark InferenceX to test Jalapeño on three open models: GPT-OSS 120B, DeepSeek R1 670B, and Kimi K2 1T.
In the DeepSeek R1 670B (MXFP4 precision) test, Jalapeño achieved approximately 1.7× throughput per watt (19,641 vs. GB300's 11,781), and end-to-end latency was reduced from 5.99 seconds to 1.65 seconds.
The Jalapeño benchmark simultaneously ran three models employing different MoE and attention mechanisms: GPT-OSS 120B, DeepSeek R1, and Kimi K2.5.
8 more timeline events
All Facts (20)
GPT-4和Mixtral等主流大模型也采用了MoE架构
85%VerifiedDeepSeek R1 does not support Function Calling by default, whereas DeepSeek V3 supports Function Calling.
80%VerifiedDeepSeek R1 offers multiple distilled versions ranging from 1.5B to 671B parameters.
80%VerifiedDeepSeek R1 is an open-source reasoning model that supports free commercial use.
80%VerifiedDeepSeek was incubated by quantitative private equity giant High-Flyer, and its R1 series models achieve reasoning capabilities close to GPT-4 level at extremely low training costs.
80%VerifiedOpenAI has introduced a 'Thinking Budget' mechanism in its o-series reasoning models, which can dynamically control the length of the internal reasoning chain.
75%VerifiedReasoning models have built-in autonomous decomposition and verification mechanisms, and overly detailed prompts can actually compress the model's reasoning space.
75%VerifiedDeepSeek R1 is released under the MIT open-source license.
65%VerifiedThe full version of DeepSeek R1 has 671B (671 billion) parameters, adopts a MoE (Mixture of Experts) architecture, and activates approximately 37B parameters during each inference.
65%UnverifiedDeepSeek经历了从V2到V3再到R1的模型迭代，模型能力和交互方式都在快速变化
95%UnverifiedDeepSeek R1是完全开源的
90%UnverifiedDeep Research Web UI支持用户自由选择LLM后端，包括DeepSeek R1和GPT-4等
90%Unverified在AIME 2024和CodeForces等权威基准测试中，DeepSeek R1的表现与OpenAI O1正式版不相上下
85%UnverifiedDeepSeek R1等具备推理能力的模型在做AI Agent时，其思维链本身就在进行类似Tree Search的规划
80%Unverified对推理模型（如DeepSeek R1）应采用简洁直接的指令，减少角色设定，不需要一步步拆解任务步骤
80%UnverifiedOllama分发的DeepSeek R1各版本均经过GGUF格式量化处理（通常为Q4_K_M或Q8_0精度），将每个参数从16位浮点数压缩至4-8位，显存占用减少50%-75%，推理质量损失通常在5%以内
80%UnverifiedGPT-OSS采用了比DeepSeek R1更严格的无监督COT（Chain-of-Thought）对齐方法
75%UnverifiedDeepSeek发布R1模型时，NVIDIA一夜蒸发6000亿美元市值，整个科技板块损失超万亿
70%UnverifiedDeepSeek R1的发布将高性能推理模型的API价格拉低至此前的数十分之一
70%Unverified推理型模型（如DeepSeek R1）强制要求格式化输出会干扰其内部推理过程导致性能下降，这是R1不支持Function Calling和JSON Output的根本原因
70%