Skip to main content
​
SERV ReasoningLaunch

​
New features

  • Claude Sonnet 5.5 available through SERV. Claude Sonnet 5.5 (claude-sonnet-5.5) costs $2.40 input / $12.00 output per million tokens, with a 1M context window. Browse models →

​
Updates

  • Staking rewards plan updated. Upcoming $SERV staking will reward stakers with a share of tokenized startups built on SERV Reasoning and launched on the platform, instead of a share of platform fees. No lockups and proportional accrual are unchanged. Read about staking →
​
SERV Reasoning

​
New features

  • GPT-6 Luna and GPT-6 Sol available through SERV. GPT-6 Luna (gpt-6-luna) costs $0.130 input / $0.650 output per million tokens, and GPT-6 Sol (gpt-6-sol) costs $2.60 / $13.00. Both have a 1M context window. Browse models →

​
Updates

  • GPT-5.6 Sol price reduced. GPT-5.6 Sol dropped to $5.25 input / $25.00 output per million tokens (from $6.50 / $40.00). The context window is unchanged. Browse models →
​
SERV ReasoningSERV Token

​
Updates

  • Gemini 2.5 Flash back in the catalog. Gemini 2.5 Flash (gemini-2.5-flash) is again available through SERV at $0.400 input / $3.50 output per million tokens, with a 1M context window. Browse models →
  • SERV token issuer details. The SERV token page now identifies OpenServ Inc. as the multisig signer entity and lists its registered address in Panama, so holders have a clear line to the managing entity. View the details →
​
SERV Reasoning

​
New features

  • New models available through SERV. Added GPT 6 Astra, Claude Opus 5, Claude Fable 5.1, and Gemini 3.1 Flash Lite, 3.5 Flash Lite, 3.6 Flash, 3.7 Flash, and 3.8 Flash. All are callable through the same SERV endpoint with pricing and context windows on the models page. Browse models →

​
Updates

  • Model catalog trimmed. Removed Gemini 2.5 Flash, Gemini 2.5 Flash Lite, Gemma 4 26B, Gemma 4 31B, Grok 4.3, Grok 4.20, Grok 4.5, DeepSeek V4 Flash, DeepSeek V4 Pro, Kimi K2.6, Kimi K2.7 Code, OpenRouter Fusion, and Z.ai GLM 5.2. If you were calling any of these API IDs, switch to another model in the catalog. Browse models →
​
SERV Reasoning

​
New features

  • SERV Reasoning is now publicly available. The private beta is over. Anyone can create an account and start making requests, with no waitlist. Read the overview →
  • Guided tutorials. A new step-by-step tutorial series takes you from account creation to a monitored production integration, covering your first request, Prompt Guard, Shadow Agent, structured outputs, streaming, prompt caching, and migration. Start the tutorials →

​
Updates

  • New blog post: The Reasoning Problem. Why free-form chain of thought is the wrong medium for reasoning, and what a bounded reasoning graph does differently. Read the post →
  • Nemotron 3 Ultra removed from the catalog. The model is no longer available through SERV. If you were calling its API ID, switch to another model in the catalog. Browse models →
​
SERV Reasoning

​
Updates

  • Moonshot AI models now listed under Kimi. Kimi models in the catalog are grouped under the Kimi provider name instead of Moonshotai. API IDs, pricing, and context windows are unchanged, so no action is needed. Browse models →
​
SERV Reasoning

​
Updates

  • Qwen models removed from the catalog. Qwen 3.7 Max, Qwen 3.7 Plus, Qwen3.6 Flash, and Qwen3.6 Max Preview are no longer available through SERV. If you were calling any of these API IDs, switch to another model in the catalog. Browse models →
​
SERV Reasoning

​
Updates

  • GPT-5.6 Luna and Terra prices reduced. GPT-5.6 Luna dropped to $0.250 input / $1.50 output per million tokens (from $1.30 / $7.80), and GPT-5.6 Terra dropped to $2.50 / $15.00 (from $3.25 / $19.50). Context windows are unchanged. Browse models →
​
SERV Token

​
Updates

  • Treasury addresses published. The SERV token page now lists the three treasury wallet addresses alongside the existing multisig disclosure, so holders can verify on-chain balances directly. View the addresses →
​
SERV Reasoning

​
Updates

  • Model selection guidance, generalized. Day One now recommends starting with the smallest, least expensive model that plausibly fits your task — rather than naming specific tiers — and moving up only when evaluations show a meaningful quality gap. Open Day One →
  • Model catalog refresh. Claude Opus 4.8 Fast has been removed from the catalog. Remaining Opus 4.x models continue to be available. Browse models →
​
SERV Reasoning

​
New features

  • serv_disable_content_filter. SERV runs a system-prompt content filter on every request by default; declare this tool by name to turn it off for requests where your application intends the model to quote or explain its own instructions. See usage →

​
Updates

  • New models available through SERV. Added Claude Opus 4.7, Opus 4.8, Opus 4.8 Fast, Sonnet 5, and Fable 5; GPT-5.6 Luna, Sol, and Terra; Gemini 2.5 Flash, 2.5 Flash Lite, 3 Flash Preview, 3.1 Pro Preview, and 3.5 Flash; Grok 4.5; Qwen 3.7 Max and 3.7 Plus; Kimi K2.6 and K2.7 Code; Nemotron 3 Ultra and OpenRouter Fusion; and Z.ai GLM 5.2. Retired the OpenAI o3, o3 Mini, o3 Pro, and o4 Mini entries. Browse the full catalog →
  • Auto-updating model catalog. The models page is now generated from the live SERV API, so pricing and context windows stay current without a docs release. Several existing models (Claude Opus 4.6, Sonnet 4.6, Grok 4.3/4.20, Qwen3.6 Flash, DeepSeek V4, GPT-5.4 Nano) now show corrected context windows.
​
SERV Reasoning

​
New features

  • SERV Tools. Enable server-side features by declaring specially named tools in any request — no SERV-specific API required. SERV detects tools prefixed with serv_, activates the feature, and strips the tool before the model runs. Works identically across the OpenAI SDK, Anthropic SDK, Vercel AI SDK, and raw HTTP. Read the reference →
  • serv_prompt_guard. Opt-in protection against prompt-injection attacks that try to leak or override your system prompt. Declare the tool by name to enable it — no parameters needed. See usage →
  • serv_shadow_agent. Runs a validate-and-iterate loop over the model’s output to raise accuracy on hard tasks. Configure hint and max_iterations through schema defaults. See usage →
​
BuildLaunchRunSERV Reasoning

​
New features

  • Build, Launch, Run. The full OpenServ platform is now organized around three product surfaces: Build for shipping agents and apps, Launch for token launches without presales or VCs, and Run for the AI Cofounder Suite that handles post-launch operations.
  • OpenClaw. A new OS-level gateway for AI agents across WhatsApp, Telegram, Discord, iMessage, and more — send a message, get an agent response. Includes a quickstart, ERC-8004 identity, x402 marketplace agents, and Telegram and Twitter integrations. Start with OpenClaw →
  • OpenServ Skills + ClawHub. Official OpenServ skills (openserv-agent-sdk, openserv-client, openserv-multi-agent-workflows, openserv-ideaboard-api, openserv-launch) are now installable into any coding agent or IDE, with ClawHub as the public registry. Browse skills →
  • No-code path. A guided no-code experience for building with Workflows, Agents, and Connect — no SDK required. Open the no-code quickstart →

​
Updates

  • SDK Integration reference. New page covering the three SERV endpoints (/v1/chat/completions, /v1/responses, /v1/messages) with side-by-side OpenAI and Anthropic SDK examples and a full parameter map. Open the reference →
  • SDK Migration guide. Step-by-step migration paths for Python (openai, anthropic), the Vercel AI SDK, LangChain, LlamaIndex, Mastra, AutoGen, CrewAI, Instructor, LiteLLM, and raw fetch. Read the guide →
  • Day One with SERV. Production-ready defaults — which model size to start with, when to upgrade, and which tasks to delegate to the model vs. your application. Open Day One →
​
SERV Reasoning

​
Updates

  • Expanded model catalog. Added GPT-5.5, GPT-5.4, GPT-5.4 Mini, GPT-5.4 Nano, Claude Opus 4.6, Claude Sonnet 4.6, Claude Haiku 4.5, Gemini Flash Latest, Gemini Pro Latest, Gemma 4, Grok 4.3 and 4.20, Qwen 3.6, and DeepSeek v4 — all available through the same SERV endpoint with pricing and context windows on a single page. Browse models →
  • Same-model performance comparison. New benchmark visual on the SERV Reasoning overview shows accuracy vs. inference cost for each model with and without SERV Reasoning on a DeFi trade-decision benchmark.
  • Quickstart, refined. First-request examples now cover the OpenAI SDK, Anthropic SDK, and raw HTTP against /v1/chat/completions, /v1/responses, and /v1/messages, with the SERV-specific behaviors called out up front. Open the quickstart →
  • reasoning_effort values clarified. Accepted values are none, low, medium, and high. If you were passing minimal, switch to low.
​
SERV Reasoning

​
New features

  • Playground. Try SERV Reasoning side by side with any base model — same prompt, two outputs, real token, cost, and latency numbers. No SDK setup required. Open the Playground →
  • OpenAI- and Anthropic-compatible inference API. Point your existing OpenAI or Anthropic SDK at the SERV endpoint and keep your prompts, tool definitions, and application logic unchanged. See the Chat Completions, Responses, and Messages references, plus the compatibility notes.
  • BRAID research framework. Published the research behind SERV Reasoning — bounded, machine-readable reasoning graphs that replace free-form chain-of-thought. Read the summary →
  • Public roadmap. Published the initial SERV Reasoning roadmap, covering the path from early access to public release, fine-tuned models, and longer-horizon research. See the roadmap →

​
Updates

  • Model catalog and pricing. Refreshed pricing for Claude Opus and Qwen models, plus a single page covering every available model, API ID, context window, and per-million-token rate. Browse models →
  • Quickstart. Streamlined the SERV Reasoning quickstart with first-request examples for the OpenAI SDK, Anthropic SDK, and raw HTTP.