Hermes Roundup: MISSINGNO Drops, Memory Deadline Looms
MISSINGNO, Nous Research's token-efficient coding and agentic model, launched as the week's headline with strong community buzz and limited free...

Created by Lou Plummer
Hermes model releases, benchmarks, architectures, and open‑source tools, agents, integrations
Explore the latest content tracked by Hermes Model & Ecosystem Tracker
MISSINGNO, Nous Research's token-efficient coding and agentic model, launched as the week's headline with strong community buzz and limited free...
Nous Research's $90 million Series B at a $1.5 billion valuation, backed by Nvidia, Microsoft M12 and others, underscores investor appetite for its...
Cursor MCP connects Hermes agents directly to your Cursor account for structured access to Cloud Agents, repositories, conversations, API keys, and...
A 100% open-source model now gives Hermes agents free live-web access while handling coding, agentic, and tool-use tasks on just 2GB RAM. This dramatically lowers hardware and cost barriers for open-source Hermes agent experimentation.
Hermes Business shifts the open-source Hermes Agent to a hosted team service, offering one central balance with per-member spend caps and a shared...
StepFun's 600B/27B MoE model is available free for one week on Nous Portal, posting agent benchmark scores in GPT-6 Luna territory. The move broadens portal access to third-party frontier models for testing and ecosystem evaluation.
Hermes and Codex differ in models, context construction, tools, and session decisions, making the prospect of shared memory a notable interoperability question.
$90M raise plus 24M downloads marks the shift from open-source Hermes Agent to 'Hermes for Businesses'.
Unanswered questions remain on:
Nous Research’s $90 million Series B at a $1.5 billion valuation, backed by NVIDIA, Microsoft’s M12, Samsung and others, directs capital to Hermes for...
HERMES replaces monolithic agents with Dev-Primitives that activate only task-critical modules via dependency graphs, directly cutting context...
Nous Research's new leaderboard moves agent assessment beyond raw accuracy to reproducible harness performance paired with real dollar cost per...
ClawBenchPro's 1k-sample, 43-domain benchmark shows agent harnesses expose weaknesses beyond raw model power, with frontier LLMs clustering at 60-74%...
Hermes connects to ChatbotX's MCP server for autonomous customer triage, tagging, and multi-channel flows across WhatsApp, Messenger, and others.
-...
Hermes users on Portal accounts now get fast, free web_search via Perplexity's agent-optimized API—no key required, ~160-230 ms latency.
Hermes Agent memory problems often stem from confusing four distinct layers rather than actual corruption.
Hermes 4 405B offers hybrid reasoning on the Llama-3.1-405B base, switching between internal <think> deliberation for depth or direct responses for speed. It also supports structured outputs, JSON mode, function calling, and tool use.
TraceDance builds custom benchmarks directly from real deployment traces to test user-specified undesirable agent behaviors that static suites...