Skip to main content

​
📅 August 2025

​
🧠 DeepSeek V3.1 Mixed Reasoning Mode Launch

August 26, 2025
DeepSeek V3.1-250821 - Revolutionary Mixed Reasoning Mode
  • 🚀 New Model: deepseek-v3-1-250821
  • 💰 Pricing Details:
    • Prompt price: $0.50 / 1M tokens (Official $0.56)
    • Completion price: $1.50 / 1M tokens (Official $1.68)
    • Save about 11% compared to official prices
    • Flexible billing for dual mode calls
  • 🧠 Core Features:
    • Think Mode: Deep reasoning with full thought process display
    • Non-Think Mode: Quick response for daily tasks
    • 128K ultra-long context window
    • Anthropic API format compatible
  • 🌟 Technical Highlights:
    • 840 billion token continuous pre-training optimization
    • New tokenizer with improved encoding efficiency
    • Beta Function Calling tool invocation support
    • Enhanced agent and tool usage capabilities
  • ⚡ Mode Switching:
    • deepseek-chat - Non-Think fast mode
    • deepseek-reasoner - Think reasoning mode
    • Official site supports “DeepThink” one-click switch
DeepSeek V3.1 is the first large model supporting mixed reasoning mode, allowing users to flexibly choose reasoning depth based on task complexity, balancing efficiency and quality.

​
🤖 Kimi K2 Official Version Integration

August 25, 2025
Kimi K2-250711 - Volcano Engine Official Partnership Version
  • 🚀 New Model: kimi-k2-250711
  • 🔄 Replaced Model: Replaces kimi-k2-0711-preview preview version
  • 🤝 Official Partnership:
    • Volcano Engine official authorized partnership version
    • Direct connection to official interface, significantly improved stability
    • Goodbye to preview version instability
  • 💰 Pricing Details:
    • Continues to maintain 15% lower than official price advantage
    • Actual charges follow the real-time console price
    • Official version performance at preview version price
  • 🌟 Core Features:
    • Excellent Chinese understanding and generation capabilities
    • Strong long text processing ability
    • More stable and reliable API responses
    • Fully compatible with original calling methods
Kimi K2-250711 is the official version in partnership with Volcano Engine, with significant improvements in stability and performance compared to preview version. Users are recommended to switch in time.

​
🚀 GPT-5 Full Series Launch

August 8, 2025
GPT-5 Full Series Official Release - OpenAI’s Strongest Model
  • 🎯 Full Series Models:
    • gpt-5 - Flagship version
    • gpt-5-2025-08-07 - Specific version
    • gpt-5-chat-latest - Latest conversation version
    • gpt-5-mini / gpt-5-mini-2025-08-07 - Lightweight version
    • gpt-5-nano / gpt-5-nano-2025-08-07 - Ultra-lightweight version
  • 💰 Pricing Details:
    • Exactly same price as OpenAI official site
    • Actual charges follow the real-time console price
    • Prompt price: $1.25 - $0.05 / 1M tokens
    • Completion price: $10.00 - $0.40 / 1M tokens
  • 🛠️ Technical Features:
    • Supports official documentation /v1/responses endpoint calls
    • Temperature parameter must be set to 1 or not passed (official restriction)
    • Use max_completion_tokens instead of max_tokens
    • gpt-5-chat-latest called through /v1/chat/completions
  • ⚡ Performance Highlights:
    • Comprehensive reasoning capabilities surpassing GPT-4 series
    • Stronger context understanding and long text processing
    • Significantly improved code generation quality
    • Multilingual capabilities reaching new heights
Important: GPT-5 series models have specific parameter requirements. Please configure according to the above technical features to ensure normal calls.

​
🎨 Claude Opus 4.1 Performance Upgrade Launch

August 7, 2025
Claude Opus 4.1 - Code Capability Breakthrough
  • 🚀 New Model: claude-opus-4-1-20250805
  • 💰 Increased Capacity Without Price Increase:
    • Price exactly same as Anthropic official site
    • Pay-as-you-go usage for better cost control
    • Same price, stronger performance
  • 🌟 Core Upgrades:
    • SWE-bench Verified reaches 74.5%, significantly improved coding ability
    • Compared to Opus 4, reasoning ability and code generation quality greatly enhanced
    • Maintains 200K ultra-long context window
    • Further optimized multilingual understanding and generation
  • ⚡ Performance Highlights:
    • Code debugging and fixing capabilities leading the industry
    • Complex reasoning task accuracy improved by 15%+
    • API response speed maintains industry-leading level
    • Perfectly compatible with all Claude series features
Claude Opus 4.1 is an iterative optimization based on Opus 4, maintaining the original powerful capabilities while making special optimizations for code generation and reasoning tasks.

​
🎯 OpenAI Open Source Models Official Launch

August 6, 2025
OpenAI’s First Open Source Large Model Release
  • 🚀 New Models:
    • gpt-oss-120b - 117 billion parameters, comparable to o4-mini performance
    • gpt-oss-20b - 21 billion parameters, can run on edge devices
  • 💰 Pricing Details:
    • Priced significantly lower than DeepSeek R1 and V3
    • Enterprise usage costs significantly reduced
  • 🌟 Core Features:
    • 128K ultra-long context window
    • Supports low/medium/high three-level reasoning modes
    • MoE architecture, extremely efficient inference
    • Apache 2.0 open source license
  • ⚡ Performance Highlights:
    • gpt-oss-120b only needs a single 80GB GPU to run
    • gpt-oss-20b supports deployment on 16GB memory devices
    • Excellent performance in coding, math, tool calling tasks
This is OpenAI’s first open source model release since GPT-2, marking an important milestone in the AI open ecosystem.
Back to the June–August 2025 archive index