- 29
- May
Claude Opus 4.8 is Anthropic's newest frontier model, released on May 28, 2026 and available everywhere immediately as the default model for premium users across Claude.ai, Claude Cowork, and Claude Code. Pricing is unchanged at $5/$25 per 1M tokens, but it is stronger at coding, long-running agentic work, and complex knowledge tasks.
What developers are talking about most: Dynamic Workflows in Claude Code, which fan a job out across tens to hundreds of parallel subagents and self-check before returning; the Effort Dial next to the model selector; and a Fast Mode that is roughly 2.5x faster and 3x cheaper than before. On top of that, Opus 4.8 is 4x less likely to let code flaws slip through unnoticed — a meaningful win for code review and security.
Claude Opus 4.8 by the Numbers
Contents
What Is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic's largest frontier model, building on Claude Opus 4.7 with improvements in three core areas — coding, long-running agentic work, and complex knowledge tasks. Anthropic states the model performs at the frontier across all three at once.
Unlike previous point releases, this one shipped tooling alongside the model: Dynamic Workflows in Claude Code, an Effort Dial for choosing how hard the model thinks in chat, and a much cheaper Fast Mode. That makes Opus 4.8 not just "a smarter model" but a step up in how AI agents are orchestrated end to end.
In short: Claude Opus 4.8 = Anthropic's top model, stronger at coding and long-running agentic work, 4x less likely to miss code flaws, and shipping with Dynamic Workflows, an Effort Dial, and a Fast Mode that is 3x cheaper — all at the same per-token price.
Opus 4.8 Basics
| Item | Details |
|---|---|
| Release date | May 28, 2026 |
| Model ID (API) | claude-opus-4-8 |
| Pricing (standard) | $5 / 1M input tokens, $25 / 1M output tokens (same as Opus 4.7) |
| Pricing (Fast Mode) | $10 / 1M input, $50 / 1M output — via usage credits |
| Default model for | premium users on Claude.ai, Cowork, and Claude Code |
| Headline new features | Dynamic Workflows, Effort Dial, 3x cheaper Fast Mode |
What's New vs. Opus 4.7
If your organization already runs Opus 4.7, the changes in 4.8 fall into two buckets — model improvements and new tooling.
| Area | Opus 4.7 | Opus 4.8 |
|---|---|---|
| SWE-bench Pro (coding) | 64.3% | 69.2% |
| Honesty / judgment | Good | 4x less likely to let code flaws pass |
| Tool calling | Standard | Fewer steps, more efficient |
| Effort control | Effort levels (e.g. xhigh) | Effort Dial next to the model selector in chat |
| Claude Code | Single subagent + /ultrareview | Dynamic Workflows — tens to hundreds of parallel subagents |
| Fast Mode | Available but expensive | ~2.5x faster, 3x cheaper ($10/$50) |
| Messages API | New system entry mid-task breaks the cache | Accepts a system entry mid-task without breaking prompt cache |
Benchmarks — Coding and Computer Use
Per Anthropic's announcement, Opus 4.8 leads on agentic coding and computer use in particular:
- SWE-bench Pro 69.2% — agentic coding, up from 64.3% on Opus 4.7 (Anthropic says it leads comparable models)
- OSWorld-Verified 84.6% — controlling a real operating system / desktop
- Online-Mind2Web 84% — browser-agent tasks operating the web on your behalf, a clear jump over the prior version
- The only model to complete every case end-to-end on the Super-Agent benchmark
But the number that matters most for enterprise work isn't a benchmark — it's honesty. Anthropic states Opus 4.8 is 4x less likely to let code flaws pass unnoticed versus the previous version, which is critical when AI helps review code headed for production.
Why "4x fewer missed flaws" matters
A model that is "too agreeable" tells you code is fine when it isn't, giving teams false confidence. Opus 4.8 pushing back more often reduces the risk that a security vulnerability or bug reaches production — especially for organizations using AI to write or review back-office systems like ERP.
Dynamic Workflows in Claude Code
Dynamic Workflows is the standout feature for developers — it lets Claude Code take a large job, fan it out across tens to hundreds of subagents running in parallel, each handling a slice, and self-check before merging results back for human review.
It suits jobs too big for a single context — a full-project migration, an audit across an entire repo, or a refactor spread over hundreds of files. The idea builds on Managed Agents and Claude Code, but makes coordinating many subagents far more systematic.
| Task type | How Dynamic Workflows helps |
|---|---|
| Full-project migration | Split into files/modules, transform in parallel, then verify each |
| Repo-wide audit / code review | Find issues per dimension (bug, security, performance), then verify each finding |
| Multi-source research | Search many angles at once, then synthesize into one report |
| High-confidence work | Have multiple agents cross-check (adversarial verify) before concluding |
A cost caveat
Because Dynamic Workflows runs many subagents at once, token usage is significantly higher than a typical session. Anthropic shipped the feature as a research preview and enabled it by default — teams managing a budget should make sure developers understand that a single workflow can cost several times an ordinary request, and reserve it for jobs that genuinely benefit from the fan-out.
The Effort Dial
The Effort Dial is a new control next to the model selector in chat and Claude Cowork, letting you choose how hard Claude thinks on each response — trading off quality, speed, and token usage.
| Effort level | Best for | Token usage |
|---|---|---|
| High (default) | Best overall balance of quality and speed for everyday work | Similar to Opus 4.7 |
| Extra-high | Difficult tasks or longer async workflows | Higher |
| Lower effort | Routine tasks where you want to stretch usage | Lower — goes further |
The default is High, which Anthropic recommends as the best balance; higher levels suit deep reasoning or long-running async jobs.
Fast Mode 3x Cheaper + Opus 4.6 Retiring
Fast Mode delivers responses roughly 2.5x faster at the same intelligence as standard Opus 4.8, priced at 2x the standard rate ($10 / 1M input and $50 / 1M output via usage credits). Anthropic says that is a 3x reduction from what Fast Mode cost before — ideal for latency-sensitive work like interactive pair-programming.
Heads-up for teams using Claude Code
Fast Mode for Opus 4.6 retires on June 29, 2026. After that, requests pinned to fast Opus 4.6 automatically fall back to standard Opus 4.6 (no errors thrown, just slower). Teams with workflows or configs pinned to fast Opus 4.6 should plan to move to Fast Mode for Opus 4.8 before the deadline.
Other extras shipping alongside
Beyond the model and the three headline features, there are extras IT teams should know about: plugin support in Claude on the web and additional Anthropic-provided plugin marketplaces (such as legal and finance), which make it easier to provision a curated set of skills to each group of users — the same modular idea you can apply to enterprise systems.
Impact on Thai Organizations + ERP
Opus 4.8's stronger agentic skills — especially computer use and fan-out workflows — directly affect how AI assists ERP work, because AI agents can operate across modules more autonomously.
| ERP Module | What AI agents can help with | Benefit |
|---|---|---|
| Procurement | Compare prices, verify terms, draft purchasing documents | Less repetitive manual work |
| Accounting | Detect anomalies, assist reconciliation, draft financial reports | Fewer errors + faster closing |
| Development / Ops | Code review that misses fewer bugs, repo-wide security audits | Safer code in production |
| Executive reporting | Pull data across modules, compile dashboards with recommendations | Faster decisions with complete data |
Saeree ERP Is Developing an AI Assistant
Saeree ERP by Grand Linux Solution Co., Ltd. is developing an AI Assistant to help users work with the ERP system more easily — aligned with the agentic-workflow direction Opus 4.8 is pushing toward an industry standard. It is currently in development and not yet generally available. Follow our progress by contacting our team.
Should You Use Opus 4.8?
| Good fit for | Consider first |
|---|---|
| Teams using AI to write/review production code | Token budget — Dynamic Workflows is token-heavy |
| Large agentic jobs (migration, audit, repo-wide refactor) | Workflows is a research preview — test before relying on it |
| Latency-sensitive work (Fast Mode now 3x cheaper) | Fast Mode still bills at 2x the standard rate |
| Organizations already on premium seats (it's the default) | Configs pinned to fast Opus 4.6 must migrate before June 29, 2026 |
"Opus 4.8 isn't just smarter — it pushes back more and works better as a team: 4x fewer missed code flaws, and the ability to fan out across hundreds of agents while checking its own work. Organizations that get their back-office systems ready for agentic AI will lead the next era."
- Saeree ERP Team
If your organization needs an ERP system ready to support AI agents in the future — contact the Saeree ERP team for a consultation.
