Model guide
Claude Opus 5.5 for Hermes Agent and OpenClaw
Anthropic's first Claude 5.5 model runs at Fable 5.1 level for 40% less than Opus 5. How to select it on a managed Hermes or OpenClaw instance, which provider row to pick on each, and the self-hosted config for both.
Reach Opus 5.5 from a managed instance
Anthropic released Claude Opus 5.5 on September 22, 2026, the first model in the Claude 5.5 family. You can move an existing Hermes Agent or OpenClaw instance onto it from the instance card, with no new instance required. The steps are the same on both frameworks — but which provider row you should pick is not, and that choice decides which account pays, so read the next section before you choose.
- Open the dashboard and find your running instance.
- Click the current model in the Text or Model row on the instance card.
- Pick the provider row that matches your framework: Anthropic on Hermes if you have saved your own Anthropic key, otherwise OpenRouter. The next section explains why, and which account each route bills.
- Search for opus 5.5. It appears as
anthropic/claude-opus-5.5on the OpenRouter row, and as Claude Opus 5.5 on the Anthropic row. The search covers the full catalogue, not only the featured shortlist, so use the search box rather than scanning the highlighted models. - Wait for the switch to finish and the instance to return to ready. Confirm the new name on the card, and start a fresh conversation if an older session is still pinned to its previous model.
Full-catalogue access applies to eligible paid instances; trial accounts use a restricted model list. If account credits run out, top up or add your own OpenRouter key on the API keys page. With a personal key in place, the tokens are billed to that OpenRouter account rather than becoming free usage.
Which provider row to pick
The answer differs by framework, and on Hermes it also depends on whether you have saved an Anthropic key. This is what decides which account is billed, so it is worth a minute.
- Hermes, with your own Anthropic key saved. Pick the Anthropic row. Hermes sends an
anthropic/selection straight toapi.anthropic.comas the hyphenated idclaude-opus-5-5, charged to your own Anthropic key. This is the most direct route, and it does not wait on our catalogue being refreshed — it reached Opus 5.5 on release day. - Hermes, with no Anthropic key saved. The same selection falls through to OpenRouter and spends OpenRouter credit instead. That works, but pick the OpenRouter row deliberately so the billing is not a surprise.
- OpenClaw. Use the OpenRouter row. An OpenClaw container resolves a model against the catalogue its own build declares, and the image OpenClaw Launch currently pins knows
claude-opus-5with no definition forclaude-opus-5-5, because the model shipped after that image was built. The OpenRouter route carries no such dependency.
True on either row: Opus 5.5 is not in our featured Anthropic shortlist yet, so search the full catalogue rather than scanning the highlighted models. Self-hosted installations are a separate matter, covered below.
What changed in Opus 5.5
Anthropic positions Opus 5.5 as performing at the level of Claude Fable 5.1 on most work while costing 40% less to run than Opus 5, with output generated more than 30% faster. Its own announcement reports these results:
| Benchmark | Opus 5.5 | Fable 5.1 | Opus 5 | GPT-6 Astra |
|---|---|---|---|---|
| Terminal-Bench 4.0 (agentic coding) | 66.4% | 55.8% | 52.3% | 57.9% |
| FrontierCode v1.1 (main) | 54.4% | 50.3% | 48.0% | 53.3% |
| CursorBench 4.0 | 57.8% | 51.8% | 46.6% | — |
| GDPval-AA v2.1 (knowledge work, Elo) | 1846 | 1735 | 1708 | 1542 |
| OSWorld 2.0 (computer use, partial) | 81.8% | 80.7% | 74.0% | — |
| AutomationBench (business workflows) | 40.0% | 31.4% | 26.9% | 41.4% |
| Terminal-Bench-Science 0.1 | 58.7% | 52.6% | 29.0% | 64.6% |
Read the last two rows as carefully as the first. GPT-6 Astra is ahead on AutomationBench and well ahead on Terminal-Bench-Science, so this is not a clean sweep. Anthropic's own framing is the useful caveat: at this level "benchmark margins have become a less reliable guide to real-world differences", and it says the gap between Opus 5.5 and Fable 5.1 in day-to-day use is narrower than the table suggests. Claude results use adaptive thinking at max effort unless noted — the exception is Terminal-Bench 4.0, where the 66.4% is Opus 5.5 at xhigh effort against Astra at high, each model's own best score.
Two other changes matter for agent work. Anthropic reports Opus 5.5 as the strongest model to date on its automated behavioural audit, and more resistant to prompt injection than Opus 5 — relevant if your agent reads untrusted web pages or messages. And its writing was deliberately reworked to put conclusions first and drop jargon, which is the difference you are most likely to notice in a chat channel.
Because Opus 5.5 is comparable to Claude Mythos 5.1 in biology and cybersecurity, Anthropic deploys it with safeguards similar to Fable 5.1's, including a verification programme for life-sciences work. Claude Sonnet 5.5 and Haiku 5.5 are expected in the following weeks.
Pricing
| Prices per 1M tokens | Opus 5.5 | Opus 5 |
|---|---|---|
| Input | $4 | $5 |
| Output | $20 | $25 |
| Cache reads | $0.20 | $0.50 |
| Cache writes | $5 | $6.25 |
The cache-read cut from $0.50 to $0.20 is the line to look at for agent workloads, where cached context is the bulk of the bill. Anthropic also offers a Fast mode at up to 2.5x speed for $8 input and $40 output per million tokens. It is not first-party-only: OpenRouter serves it as a separate anthropic/fast endpoint at those same rates.
OpenRouter lists anthropic/claude-opus-5.5 at the same $4 and $20, verified September 22, 2026 (UTC), plus a batch entry at half those rates. These are token prices, not your hosting subscription. Routing and pricing can change, so check the model page before committing to a long run.
One caution on the headline saving. The 40% figure is Anthropic's measurement on typical workloads and combines a lower per-token price with fewer tokens per task. Opus 5.5 is a reasoning model with effort levels from low through max, and a high effort setting spends substantially more thinking tokens per answer. Compare on your own tasks at the effort level you actually intend to use before you treat the saving as banked.
Self-hosted Hermes Agent
On your own Hermes installation, run hermes model and select the provider and model, or set the model section of ~/.hermes/config.yaml directly. Using your own Anthropic key:
model:
provider: anthropic
default: claude-opus-5-5
The provider lives in its own key, so the model value stays bare — no anthropic/ prefix. This route needs ANTHROPIC_API_KEY in your environment or Hermes credential store. Through OpenRouter instead, the default carries OpenRouter's own slug:
model:
provider: openrouter
default: anthropic/claude-opus-5.5
Thinking effort is configured separately, and in a different section — it belongs under agent, not under model:
agent:
reasoning_effort: "medium"
reasoning_effort accepts minimal, low, medium, high, xhigh and none, and applies on the OpenRouter and Nous Portal routes. Per-model overrides go in agent.reasoning_overrides, which has no hermes config set support and must be edited in YAML.
Anthropic's Fast mode is agent.service_tier: "fast", sent only to first-party endpoints — pair it with provider: anthropic, not the OpenRouter route. A model change applies to new sessions; use hermes gateway restart if you want to restart the gateway outright. Field names are from the upstream configuration example.
Self-hosted OpenClaw
For a self-hosted OpenClaw installation, add your Anthropic key through onboarding, then set the model:
openclaw onboard --auth-choice anthropic-api-key
openclaw models set anthropic/claude-opus-5-5
The matching section of ~/.openclaw/openclaw.json is below. Merge it into your existing configuration rather than replacing the file:
{
"agents": {
"defaults": {
"model": { "primary": "anthropic/claude-opus-5-5" }
}
}
}
This requires an OpenClaw build whose Anthropic catalogue already knows claude-opus-5-5. On an older build, route through OpenRouter instead, where the id takes the extra provider prefix:
openclaw models set openrouter/anthropic/claude-opus-5.5
Note the spelling difference that catches people out: Anthropic's own API id is hyphenated (claude-opus-5-5) while OpenRouter's slug is dotted (claude-opus-5.5). Managed OpenClaw Launch customers do not edit either file — use the instance-card steps above.
Sources and related guides
- Anthropic: Introducing Claude Opus 5.5
- OpenRouter: anthropic/claude-opus-5.5
- Hermes Agent + Anthropic and OpenClaw + Anthropic
- Claude Opus 5 guide and Claude Fable 5.1 guide
- Hermes hosting and OpenClaw hosting
Frequently asked questions
Does Claude Opus 5.5 work with Hermes Agent?
Yes, and Hermes has the more direct route of the two frameworks. With your own Anthropic key saved, selecting Opus 5.5 on the Anthropic row sends it straight to api.anthropic.com as claude-opus-5-5. Without an Anthropic key it routes through OpenRouter as anthropic/claude-opus-5.5. Self-hosted Hermes takes provider: anthropic with default: claude-opus-5-5.
Why is Opus 5.5 not in the featured Anthropic model list on my dashboard?
The featured shortlist is a curated catalogue that has not been refreshed since the model launched on September 22, 2026. Opus 5.5 is still reachable today by selecting the OpenRouter provider and searching the full catalogue for opus 5.5.
Should I pick Opus 5.5 from the Anthropic provider row or the OpenRouter row?
On Hermes, if you have saved your own Anthropic key, use the Anthropic row: Hermes sends the selection direct to api.anthropic.com as claude-opus-5-5 and your own Anthropic key pays. Without a saved Anthropic key the same pick falls through to OpenRouter. On OpenClaw, use the OpenRouter row — the pinned container image has no definition for claude-opus-5-5, so an OpenClaw container cannot resolve that selection.
Is Opus 5.5 really 40% cheaper than Opus 5?
Per-token it is 20% cheaper on input and output ($4 and $20 against $5 and $25) and 60% cheaper on cache reads. Anthropic's 40% figure is a typical-workload measurement that also assumes fewer tokens per task. Higher reasoning effort spends more thinking tokens, so compare on your own tasks at the effort level you intend to use.
Do I need a new instance, or a direct Anthropic API key?
No new instance is required. The OpenRouter route needs no Anthropic key at all — it spends eligible account credits or your own OpenRouter key. An Anthropic key is not mandatory either, but on managed Hermes it changes the routing: with one saved, an Anthropic-row pick goes direct to api.anthropic.com on your key instead of through OpenRouter. For self-hosted installations a direct key is the normal route.
Will my Opus 5 instance upgrade to Opus 5.5 automatically?
No. Opus 5.5 is a separate model id. Select it on the instance card and wait for the change to finish, then start a fresh session if an existing conversation is still on the old model.