Available in Onevium today, at the same price as 4.7
Anthropic announced Claude Opus 4.8 earlier today. The headline from their post is short: it builds on Opus 4.7 with sharper judgment, more honesty about its own progress, and the ability to work independently for longer than its predecessors. Available immediately, at the same per-token price as 4.7.
Onevium 1.1.9 ships with claude-opus-4-8 already wired into the model picker, the effort tiers (low / medium / high / xhigh / max), and the usage-cost catalog. If you already have an Anthropic provider connected, a one-time DB migration (v33) reseeds your model list so 4.8 shows up under Settings → Providers → Manage Models without re-registering the provider.
There is nothing else to do. Pick claude-opus-4-8 from the model dropdown and the next chat — or scheduled task, or Channel message — runs against it.
The benchmarks — where 4.8 actually moves the needle
Headline numbers don't lie, but pricing the move requires looking at where 4.8 widens the gap. The most material jumps over 4.7 are on the agentic / multi-step benchmarks — the ones that matter most when you're letting Onevium run a long task end-to-end.
- Agentic coding (SWE-Bench Pro) — 69.2% vs 64.3% on 4.7. ~5-point absolute lift on real-world repo-modification tasks; the largest jump in this release.
- Multidisciplinary reasoning (Humanity's Last Exam) — 49.8% no-tools vs 46.9% on 4.7 (+2.9), and 57.9% with-tools vs 54.7% (+3.2). The tool-augmented number is where Onevium-style workflows live.
- Knowledge work (GDPval-AA) — 1890 vs 1753 on 4.7. The biggest absolute headroom Anthropic has published on this benchmark in a single model bump.
- Agentic terminal coding (Terminal-Bench 2.1) — 74.6% vs 66.1% on 4.7. GPT-5.5 still leads this column at 78.2%, but 4.8 closes most of the prior gap.
- Computer use (OSWorld-Verified) — 83.4% vs 82.8% on 4.7. Smaller delta, but 4.8 still leads every comparable model.
- Agentic financial analysis (Finance Agent v2) — 53.9% vs 51.5% on 4.7. Small but consistent.
What "works independently for longer" means inside Onevium
Anthropic's announcement leans hard on "works independently for longer than its predecessors." In a chat product that's a vibe; inside Onevium, where a single task can spawn background scheduled runs, long-running tool chains, and channel-bridged conversations, it's a measurable property.
Two surfaces benefit most. First, scheduled tasks — automations that fire on cron and run a multi-step prompt against your codebase. The longer a model can stay coherent, the more useful it is to point a scheduled task at "audit X subsystem and open PRs for what you find" instead of breaking it into hand-held stages. Second, the new Memory v2 window-extraction pipeline that also ships in 1.1.9 — it reasons over the last N user+assistant turns of a session, and a stronger reasoning model translates directly into denser, less-redundant memories.
If you've been bouncing between 4.7 and 4.6 for long-horizon work because 4.7 sometimes loses the thread, the practical upgrade path is: set 4.8 as your default in Settings → Providers → Anthropic, and re-run the same scheduled task you trust 4.6 with. Most users will see the iteration count drop and the quality of the final patch go up.
Pricing, effort tiers, and cost visibility
Opus 4.8 launches at exactly the same per-million-token pricing as 4.5 / 4.6 / 4.7: $5 input, $25 output, $0.50 cache-read, $6.25 5-minute cache-write. Onevium's usage-cost catalog reflects this on day one, so Settings → Activity → Total Cost stays accurate from the moment you switch models.
Effort tiers carry over from 4.7: low / medium / high / xhigh / max are all available, with `high` as the default. If you were already using xhigh or max on 4.7 for hard reasoning tasks, those exact tiers exist on 4.8 — no need to relearn the dial.
One caveat worth naming: the SDK reports actual spend for Anthropic-direct connections, so Pro and Max subscribers will continue to see $0 on the Total Cost card even after switching to 4.8. That's correct, not a bug — Onevium trusts the SDK's cost_usd field over the catalog rate for Anthropic providers, exactly because the subscription pricing diverges from the per-token rate card.
How to switch — three places
Onevium lets you pick a model in three independent places, and the 1.1.9 update wires Opus 4.8 into all of them:
- Chat composer model dropdown — top-right of the message input. Per-message override; great for trying 4.8 on a single hard prompt without changing your defaults.
- Settings → Providers → Anthropic → Manage Models — toggles which models show in the dropdown at all. claude-opus-4-8 appears automatically after the v33 migration runs (next time you launch Onevium 1.1.9).
- Settings → Models → Default model — sets the model used when you don't pick one per-message, and what scheduled tasks / Channel bots default to. Switch this to claude-opus-4-8 to get 4.8 on long-running automation.
Updating
Onevium 1.1.9 is available now for macOS (Apple Silicon and Intel), Windows, and Linux. If you're already on 1.1.6 or later, the in-app update prompt will surface within a launch or two. Otherwise the latest downloads are on the Releases page.