disler · shamrockmuffin · Aug 11, 2025
diff --git a/ai_docs/anthropic_openai_compat.md b/ai_docs/anthropic_openai_compat.md
@@ -22,6 +22,8 @@ print(response.choices[0].message.content)
 ```
 
 ## Available Models
+- claude-opus-4-20250514
+- claude-sonnet-4-20250514
 - claude-3-opus-20240229
 - claude-3-sonnet-20240229
 - claude-3-haiku-20240307
@@ -34,6 +36,8 @@ print(response.choices[0].message.content)
 3. Audio input not supported
 4. Prompt caching not supported via OpenAI SDK (use native Anthropic SDK)
 5. Rate limits follow Anthropic's standard limits
+6. Extended context windows (200K tokens) for long-horizon tasks
+7. Enhanced reasoning capabilities for complex problem-solving
 
 ## Supported Features
 - ✅ Chat completions
@@ -69,4 +73,19 @@ response = client.chat.completions.create(
 ## Important Notes
 - This compatibility layer is for testing and comparison, not production
 - For full features (PDFs, citations, prompt caching), use native Anthropic SDK
-- Most unsupported fields are silently ignored rather than producing errors
+- Most unsupported fields are silently ignored rather than producing errors
+- Claude 4 models (Opus and Sonnet) support extended context windows and enhanced reasoning capabilities
+- For the latest model availability and features, refer to Anthropic's official API documentation
+
+## Claude 4 Latest Capabilities (August 2025)
+
+### Enhanced Reasoning
+- **Long-Horizon Tasks**: Can execute complex, multi-step reasoning over extended periods
+- **Coding Proficiency**: Optimized for software development and code analysis
+- **Data Analysis**: Advanced capabilities for analyzing large datasets and complex information
+
+### Use Cases
+- **Agentic Workflows**: Configurable reasoning efforts for different complexity levels
+- **Extended Context**: 200K token context windows for comprehensive document analysis
+- **Complex Problem-Solving**: Multi-step reasoning for engineering and research tasks
+- **Code Generation**: Advanced coding assistance with understanding of large codebases
diff --git a/ai_docs/gpt_oss_ollama_guide.md b/ai_docs/gpt_oss_ollama_guide.md
@@ -4,17 +4,43 @@ Source: [OpenAI Cookbook](https://cookbook.openai.com/articles/gpt-oss/run-local
 
 ## Model Options
 
-### gpt-oss-20b
-- Smaller model
-- Best with ≥16GB VRAM or unified memory
-- Perfect for higher-end consumer GPUs or Apple Silicon Macs
-
-### gpt-oss-120b
-- Full-sized model
-- Best with ≥60GB VRAM or unified memory
-- Ideal for multi-GPU or workstation setups
-
-**Note:** Models ship MXFP4 quantized out of the box. CPU offloading is possible but slower.
+### gpt-oss:20b
+- **Release Date**: August 5, 2025
+- **License**: Apache 2.0
+- **Smaller model** optimized for consumer hardware
+- **Best with ≥16GB VRAM or unified memory**
+- **Perfect for higher-end consumer GPUs or Apple Silicon Macs**
+- **Context Length**: 131K tokens
+- **Hardware Requirements**: 40GB VRAM (FP16), ~10GB VRAM (4-bit quantized)
+
+### gpt-oss:120b
+- **Release Date**: August 5, 2025
+- **License**: Apache 2.0
+- **Full-sized model** for high-performance tasks
+- **Best with ≥80GB VRAM or unified memory**
+- **Ideal for multi-GPU or workstation setups**
+- **Context Length**: 32K tokens
+- **Hardware Requirements**: 240GB VRAM (FP16), ~60GB VRAM (4-bit quantized)
+
+**Note:** Models ship MXFP4 quantized out of the box. CPU offloading is possible but slower. Both models are now available on Amazon Bedrock and AWS SageMaker for cloud deployment.
+
+## Latest Capabilities (August 2025)
+
+### Enhanced Reasoning
+- **Configurable Reasoning Efforts**: Adjust reasoning depth for different task complexities
+- **Agentic Workflows**: Optimized for autonomous AI agents and multi-step reasoning
+- **Extended Context**: 131K tokens for gpt-oss:20b, 32K for gpt-oss:120b
+
+### Use Cases
+- **On-Device AI**: Privacy-focused applications with local processing
+- **Cost-Efficient Deployment**: Free open-weight models for production use
+- **Extended Document Analysis**: Process large documents and codebases
+- **Multi-Step Problem Solving**: Complex reasoning tasks requiring sustained attention
+
+### Cloud Integration
+- **Amazon Bedrock**: Native integration for enterprise deployment
+- **AWS SageMaker**: Pre-configured endpoints and scaling
+- **Hybrid Deployment**: Combine local Ollama with cloud services
 
 ## Quick Setup
 

diff --git a/ai_docs/new_openai_gpt_models.md b/ai_docs/new_openai_gpt_models.md
@@ -12,6 +12,7 @@ OpenAI has introduced two groundbreaking model families in August 2025:
 - **Announcement Date**: August 7, 2025
 - **Rollout**: Gradual API access to Plus subscribers
 - **Full Availability**: Rolling out across platforms
+- **Latest Update**: August 11, 2025 - General availability expanding
 
 ### Model IDs for API Access
 
@@ -25,6 +26,8 @@ OpenAI has introduced two groundbreaking model families in August 2025:
 | Chat-optimized | `gpt-5-chat-latest` | Non-reasoning, fast streaming |
 | **Aliases** | | |
 | Latest stable | `gpt-5-2025-08-06` | Date-versioned stable release |
+| **New Models** | | |
+| Deep Research | `gpt-deep-research` | AI agent for extensive web browsing and analysis |
 
 ### Model Variants
 
@@ -139,8 +142,9 @@ chat_response = client.chat.completions.create(
 
 ### Overview
 - **Announcement Date**: August 5, 2025
-- **Type**: Open-weight models under MIT license
-- **Availability**: Weights available on Hugging Face
+- **Type**: Open-weight models under Apache 2.0 license
+- **Availability**: Weights available on Hugging Face and AWS
+- **Latest Update**: August 11, 2025 - Now available on Amazon Bedrock and SageMaker
 
 ### Model Variants
 
@@ -252,6 +256,7 @@ trainer.train()
 
 #### Cloud Deployment
 - **AWS SageMaker**: Pre-configured endpoints for gpt-oss models
+- **Amazon Bedrock**: Native integration for gpt-oss-120b and gpt-oss-20b
 - **Azure AI Foundry**: Native GPT-5 integration via Azure OpenAI Service
 - **Google Cloud Vertex AI**: Model garden integration for open models
 - **Replicate**: One-click deployment for gpt-oss variants
@@ -262,6 +267,20 @@ trainer.train()
 - **Core ML**: Apple Silicon optimization for on-device inference
 - **Mobile SDKs**: Direct integration via OpenAI mobile libraries
 
+### New Models
+
+#### gpt-deep-research
+- **Model ID**: `gpt-deep-research`
+- **Release Date**: February 2, 2025
+- **Type**: AI agent leveraging OpenAI's o3 model
+- **Capabilities**: 
+  - Extensive web browsing and data analysis
+  - Multi-step research tasks
+  - Comprehensive report generation
+  - Data synthesis and analysis
+- **Availability**: Initially ChatGPT Pro subscribers, lightweight version for all users since April 2025
+- **Use Cases**: In-depth research, data analysis, comprehensive reporting
+
 ## Comparison Matrix
 
 | Feature | GPT-4o | GPT-5 | GPT-5-mini | GPT-5-nano | gpt-oss-120b | gpt-oss-20b |

diff --git a/ai_docs/ollama_openai_compat.md b/ai_docs/ollama_openai_compat.md
@@ -30,10 +30,29 @@ print(response.choices[0].message.content)
 ## Available Models
 
 Any model you have pulled with Ollama can be used. Popular options:
+
+### Latest State-of-the-Art Models (August 2025)
+- **gpt-oss:20b**: OpenAI's open-weight model optimized for 16GB GPUs, supports 131K context length
+- **gpt-oss:120b**: OpenAI's larger model requiring 80GB GPU for enhanced capabilities
+- **mistral:7b**: Mistral AI's efficient 7B parameter model with strong performance
+- **mixtral:8x7b**: Mistral AI's mixture of experts model for robust multi-task performance
+- **mistral:large**: Mistral AI's latest large model for advanced reasoning
+- **magistral:small**: Mistral AI's first AI reasoning model with chain-of-thought capabilities (open-source)
+- **magistral:medium**: Mistral AI's commercial reasoning model for complex problem-solving
+- **gemma:2b**: Google's lightweight 2B parameter model for general text generation
+- **gemma:7b**: Google's 7B parameter model with enhanced capabilities
+- **qwen2:0.5b**: Alibaba's ultra-lightweight model for resource-constrained applications
+- **qwen2:7b**: Alibaba's 7B parameter model with improved performance
+- **llama3.2:3b**: Meta's latest 3B parameter model optimized for efficiency
+- **llama3.2:8b**: Meta's 8B parameter model with enhanced reasoning
+- **llama3.2:70b**: Meta's 70B parameter model for complex tasks
+- **llama3.3:70b**: Meta's latest 70B model emphasizing cost-efficiency and performance
+- **deepseek:r1**: DeepSeek's open-source model with strong coding and mathematics capabilities
+- **grok:3**: xAI's real-time conversational AI with up-to-date information access
+
+### Established Popular Models
 - llama2
-- mistral
 - codellama
-- mixtral
 - deepseek-coder
 - phi
 - neural-chat
@@ -42,6 +61,25 @@ Any model you have pulled with Ollama can be used. Popular options:
 
 To pull a model: `ollama pull <model_name>`
 
+**Note**: The gpt-oss models are OpenAI's latest open-weight releases, offering enterprise-grade performance on consumer hardware. Use the exact model names with colons (e.g., `gpt-oss:20b`) when pulling models.
+
+## Latest Model Capabilities (August 2025)
+
+### Reasoning and Problem-Solving
+- **Mistral Magistral Series**: First AI reasoning models with chain-of-thought capabilities for complex problem-solving
+- **GPT-OSS Models**: Enhanced reasoning with configurable effort levels for agentic workflows
+- **Llama 3.3**: Optimized for mathematical computations and instruction following
+
+### Specialized Use Cases
+- **DeepSeek R1**: Strong performance in coding assistance and mathematical problem-solving
+- **Grok 3**: Real-time conversational AI with access to current information
+- **Claude 4 Series**: Extended context windows and enhanced reasoning for long-horizon tasks
+
+### Hardware Optimization
+- **gpt-oss:20b**: Optimized for 16GB+ VRAM with 131K context support
+- **gpt-oss:120b**: High-performance model requiring 80GB+ VRAM
+- **Magistral Small**: Open-source reasoning model for accessible deployment
+
 ## API Endpoint
 
 - Base URL: `http://localhost:11434/v1`

diff --git a/ai_docs/openai_agent_sdk_core.md b/ai_docs/openai_agent_sdk_core.md
@@ -13,6 +13,25 @@ The OpenAI Agents SDK enables building production-ready agentic AI applications
 - **Sessions**: Automatic conversation history management
 - **Streaming**: Real-time progress monitoring
 
+### Latest Model Capabilities (August 2025)
+
+#### GPT-5 Series for Engineering
+- **gpt-5**: PhD-level expertise for complex architectural decisions
+- **gpt-5-mini**: 80% performance at 20% cost for production workflows
+- **gpt-5-nano**: Ultra-fast responses for CI/CD and edge deployment
+- **400K Context**: Process entire codebases in single conversations
+
+#### GPT-OSS for Local Development
+- **gpt-oss:20b**: Consumer hardware optimization (16GB+ VRAM)
+- **gpt-oss:120b**: High-performance local deployment (80GB+ VRAM)
+- **Apache 2.0 License**: Free for commercial use
+- **131K Context**: Extended reasoning for complex engineering tasks
+
+#### Claude 4 for Research
+- **200K Context**: Comprehensive codebase analysis
+- **Enhanced Reasoning**: Multi-step problem solving
+- **Long-Horizon Tasks**: Extended engineering workflows
+
 ## Installation and Setup
 
 ```bash

diff --git a/ai_docs/openai_agents_model_configuration.md b/ai_docs/openai_agents_model_configuration.md
@@ -17,6 +17,20 @@ The OpenAI Agents SDK supports multiple model providers through different config
 - Fallback for providers without Responses API support
 - More widely compatible with OpenAI-compatible endpoints
 
+### 2. Latest Model Families (August 2025)
+
+#### GPT-5 Series
+- **gpt-5**: Flagship reasoning model with PhD-level expertise
+- **gpt-5-mini**: 80% of GPT-5 capabilities at 20% cost
+- **gpt-5-nano**: Ultra-lightweight for edge deployment
+- **gpt-5-chat-latest**: Non-reasoning chat model for fast responses
+
+#### GPT-OSS Series
+- **gpt-oss:20b**: Open-weight model for consumer hardware (16GB+ VRAM)
+- **gpt-oss:120b**: High-performance open-weight model (80GB+ VRAM)
+- **License**: Apache 2.0 - Free for commercial use
+- **Availability**: Local deployment via Ollama, cloud via AWS Bedrock
+
 ## Non-OpenAI Model Integration
 
 ### Method 1: LiteLLM Integration (Simplest)
@@ -159,14 +173,17 @@ Solution: Use providers that support JSON schema outputs, or handle malformed JS
 
 ## Feature Compatibility Matrix
 
-| Feature | OpenAI | Anthropic (via LiteLLM) | Ollama |
-|---------|--------|-------------------------|---------|
-| Responses API | ✅ | ❌ | ❌ |
-| Chat Completions | ✅ | ✅ | ✅ |
-| Structured Outputs | ✅ | ⚠️ | ❌ |
-| Multimodal | ✅ | ✅ | ⚠️ |
-| Function Calling | ✅ | ✅ | ❌ |
-| File/Web Search | ✅ | ❌ | ❌ |
+| Feature | OpenAI | Anthropic (via LiteLLM) | Ollama | GPT-OSS |
+|---------|--------|-------------------------|---------|---------|
+| Responses API | ✅ | ❌ | ❌ | ❌ |
+| Chat Completions | ✅ | ✅ | ✅ | ✅ |
+| Structured Outputs | ✅ | ⚠️ | ❌ | ⚠️ |
+| Multimodal | ✅ | ✅ | ⚠️ | ❌ |
+| Function Calling | ✅ | ✅ | ❌ | ⚠️ |
+| File/Web Search | ✅ | ❌ | ❌ | ❌ |
+| Extended Context | ✅ (400K) | ✅ (200K) | ✅ (131K) | ✅ (131K) |
+| Reasoning Effort | ✅ (GPT-5) | ✅ (Claude 4) | ❌ | ⚠️ |
+| Local Deployment | ❌ | ❌ | ✅ | ✅ |
 
 ## Key Implementation Notes
 
@@ -215,4 +232,72 @@ ollama_agent = Agent(
         openai_client=ollama_client
     )
 )
+```
+
+## Integration with Latest Models
+
+### GPT-5 Reasoning Models
+
+```python
+# Using GPT-5 with reasoning capabilities
+reasoning_agent = Agent(
+    name="Research Agent",
+    model="gpt-5",  # or "gpt-5-mini" for cost optimization
+    model_settings=ModelSettings(
+        temperature=0.1,
+        extra_args={
+            "reasoning_effort": "high",  # GPT-5 specific parameter
+            "max_tokens": 128000,  # Up to 128K output tokens
+        }
+    )
+)
+
+# For edge deployment with minimal reasoning
+edge_agent = Agent(
+    name="Edge Agent",
+    model="gpt-5-nano",
+    model_settings=ModelSettings(
+        extra_args={
+            "reasoning_effort": "minimal",  # Fastest, cheapest
+        }
+    )
+)
+```
+
+### GPT-OSS Local Models
+
+```python
+# Local GPT-OSS deployment via Ollama
+from openai import AsyncOpenAI
+
+ollama_client = AsyncOpenAI(
+    base_url="http://localhost:11434/v1",
+    api_key="ollama"
+)
+
+local_agent = Agent(
+    name="Local Agent",
+    model=OpenAIChatCompletionsModel(
+        model="gpt-oss:20b",  # Use exact Ollama model name
+        openai_client=ollama_client
+    ),
+    model_settings=ModelSettings(
+        temperature=0.2,
+        max_tokens=131000,  # Full context window
+    )
+)
+```
+
+### Claude 4 Models
+
+```python
+# Claude 4 with extended context
+claude_agent = Agent(
+    name="Claude Agent",
+    model="litellm/anthropic/claude-opus-4-20250514",
+    model_settings=ModelSettings(
+        temperature=0.1,
+        max_tokens=200000,  # Claude 4's 200K context
+    )
+)
 ```