February

​

Summary

​

Highlights

• Expanding go-to-market — Meeting the growing enterprise demand across finance, pharma, technology, and beyond
• Governance for agentic AI — Building the controls organizations need as agents take autonomous action: permissions, identity, access boundaries, and budget guardrails
• Platform infrastructure at scale — Higher-volume workloads, real-time use cases, and day0 support for new models and pricing changes

​

Platform

​

All Organizations Now on Model Catalog

• Browse all available models across 40+ providers in one place
• Configure model-specific settings without touching code
• Switch providers for the same model with a single change

​

Gateway

​

Use the Responses API with Any Provider

• Keep a single API format while switching between Anthropic, Google, Bedrock, and others
• Use prompt caching and thinking parameters across providers

from portkey_ai import Portkey

portkey = Portkey(api_key="PORTKEY_API_KEY")

response = portkey.responses.create(
    model="@anthropic-provider/claude-sonnet-4-5-20250514",
    input="Explain quantum computing in simple terms"
)

print(response.output_text)

​

Use the Messages API with Any Provider

import anthropic

client = anthropic.Anthropic(
    api_key="PORTKEY_API_KEY",
    base_url="https://api.portkey.ai"
)

message = client.messages.create(
    model="@google-provider/gemini-2.5-flash",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain quantum computing in simple terms"}]
)

print(message.content[0].text)

• Keep your existing Messages API code while routing to OpenAI, Google, and more
• Let the gateway handle format conversion automatically

​

Guardrails

​

Protect Your Apps with Zscaler AI Guard

• Enforce Detection Policies for security checks
• Block or flag data loss risks with DLP protection
• Catch prompt injection attempts on both inputs and outputs

​

Models and providers

• Databricks: You can now route requests to Databricks Model Serving for chat completions, completions, and embeddings. Set up Databricks.
• Claude 4.6: Use Claude 4.6 features across Anthropic, Bedrock, and other providers — including Adaptive Thinking with reasoning_effort, structured outputs via output_config, and new stop reasons like refusal.
• Together AI reasoning: Run reasoning/thinking models on Together AI with the reasoning_effort parameter and get structured content_blocks in responses. Try Together AI reasoning.
• Bedrock Anthropic citations: Access Anthropic’s citations feature on Bedrock through the chat completions API.

​

Enhancements

• OpenAI & Azure OpenAI TTS streaming: Stream text-to-speech audio via Server-Sent Events by setting stream_format: “sse”. Set up SSE streaming.
• ZhipuAI: Generate images with ZhipuAI’s CogView models (e.g., cogview-4-250304). See ZhipuAI docs.
• Vertex AI: Control image and video input resolution with media_resolution, skip PTU cost attribution with vertex_skip_ptu_cost_attribution, and configure workload identity auth via x-portkey-vertex-auth-type.
• Batch pricing: Get accurate cost attribution for batch requests with dedicated batch pricing. When batch-specific pricing isn’t available, costs default to 50% of standard pricing.

​

Community & Events

​

Agent Harness Salon: Bangalore

​

Meet us at RSA!

​

Resources

• Blog: LLM Deployment Pipeline Explained Step by Step
• Blog: How to host an AI Hackathon without losing control of your keys or budget
• Blog: Securing the MCP Gateway: Lasso Partners with Portkey to Deliver Enterprise-Grade Agentic AI Protection
• Blog: The best approach to compare LLM outputs

​

Support