LoginSign Up
← Back to all sparks
O

opencode

AI-ASSISTANTS
Velocity5.0

Open-source AI coding agent built for the terminal

opencode.ai ↗

OpenCode ships daily with GPT-6/Astra support, Claude 5.1 thinking blocks, and Azure enterprise auth

multi-providerreasoning-modelsclaude-5gpt-6azure-integrationagentic-coding
◆Current state
OpenCode is shipping daily-to-weekly micro-releases with a tight focus on multi-provider reliability. GPT-6/Astra system prompt support, Claude 5.1 thinking block binding, and Azure Entra ID CLI sign-in all landed in the past two weeks. Each release also includes fixes for specific model providers (Bedrock DeepSeek, OpenAI compatibility, Cloudflare AI Gateway routing), reflecting the maintenance cost of maintaining a wide provider compatibility surface as that surface keeps expanding with new models.
◆Where it's heading
The pattern of adding reasoning effort variants, thinking block management, and adaptive thinking summarization across providers shows OpenCode investing in first-class support for extended reasoning models—the category where Claude 5 and GPT-6 both live. Azure authentication improvements and Cloudflare AI Gateway routing fixes suggest the enterprise integration surface is expanding alongside model breadth. Anthropic's Claude 5.1 thinking block handling is a particularly active area.
◆Prediction
Expect continued day-one support for major model releases from Anthropic, OpenAI, and AWS, with Claude 5.x and GPT-6 extended reasoning capabilities becoming better-differentiated features. Azure enterprise auth improvements are likely to continue through the current cycle.

◆Recent moves

  1. 1d ago

    v1.18.31: ACP Session Continuity and GitHub Copilot Adaptive Thinking

    Session options and reasoning boundaries now survive loading, resuming, and forking in OpenCode's ACP mode—a continuity fix that matters for long-running agentic sessions. GitHub Copilot models also gain summarized adaptive thinking output, tightening the reasoning display for that provider.

    View source ↗
  2. 7d ago

    v1.18.30: GPT-6 Astra Support and Multi-Provider SDK Updates

    GPT-6 models get an Astra system prompt, and reasoning effort variants land for GitLab GPT and Claude models. Multiple provider SDK updates (Bedrock DeepSeek, Azure, OpenAI) show the breadth of the compatibility surface OpenCode now has to maintain across new model generations.

    View source ↗
  3. 11d ago

    v1.18.29: GPT-6 Codex OAuth Filter Fix

    Fixes Codex OAuth model filtering to recognize integer GPT version IDs like gpt-6—a narrow bug that was preventing GPT-6/Astra from appearing for OpenAI subscription users. Narrow scope; no feature change.

    View source ↗
  4. 11d ago

    v1.18.28: GitHub Copilot Session Header and Desktop Auth Fixes

    Sends the session ID in GitHub Copilot's interaction header for better request tracking, and fixes desktop client authentication during device login. Internal plumbing improvements; no user-visible feature change.

    View source ↗
  5. 13d ago

    v1.18.27: Provider Timeouts Extended and Claude 5.1 Thinking Blocks Scoped

    Default provider header timeouts extend to five minutes—a practical fix for slow model cold-starts causing spurious failures. Claude 5.1 thinking block binding is now scoped to that model version specifically, preventing older Claude deployments from rejecting extended reasoning requests they don't support.

    View source ↗
  6. 14d ago

    v1.18.26: Claude 5 Session Tolerance and Azure CLI Sign-In

    Claude 5 sessions now tolerate stale thinking blocks from changed prompts or tool swaps instead of failing outright—useful for iterative sessions where context shifts mid-conversation. Azure CLI sign-in drops the dependency on Azure management APIs, simplifying enterprise setup without requiring an API key.

    View source ↗