Claude Opus 4.8 Released: How Does It Stack Up Against DeepSeek V4 Pro?
Anthropic shipped Claude Opus 4.8 today with sharper benchmarks, dynamic workflows in Claude Code, and a 3× cheaper fast mode. We line it up against DeepSeek V4 Pro on the metrics that matter.
Key Takeaways
- • Claude Opus 4.8 launched May 28, 2026 — same price as Opus 4.7, broadly better benchmarks.
- • New "effort control" on claude.ai lets users dial how hard Claude thinks per task.
- • Dynamic workflows in Claude Code targets very large-scale autonomous engineering tasks.
- • Fast mode for Opus 4.8 runs at 2.5× speed and is now 3× cheaper than on previous Opus versions.
- • DeepSeek V4 Pro remains 17–86× cheaper per token — the value gap is unchanged.
What's New in Claude Opus 4.8
Anthropic positioned Opus 4.8 as a steady, incremental upgrade over Opus 4.7 rather than a generational leap. Early testers describe it as "more reliable and sharper in its judgement" on agentic tasks — meaning fewer dead-end tool calls, better recovery from errors, and stronger long-horizon planning.
Three concrete changes shipped alongside the model:
- Effort control on claude.ai — a user-facing slider for how much compute Claude burns per request.
- Dynamic workflows inside Claude Code — designed for very large-scale, multi-file engineering problems that previously needed manual orchestration.
- Fast mode running at 2.5× the speed of standard Opus, now priced 3× lower than the previous fast tier.
Source: Anthropic announcement, May 28, 2026.
Benchmarks: Opus 4.8 vs DeepSeek V4 Pro
Anthropic's published table shows Opus 4.8 leading Opus 4.7 across coding, agentic and reasoning evaluations. DeepSeek hasn't re-benchmarked V4 Pro against the 4.8 release yet, but based on the May 1, 2026 V4 preview numbers the picture looks like this:
| Benchmark | Claude Opus 4.8 | DeepSeek V4 Pro |
|---|---|---|
| SWE-bench Verified | ~82% | >80% |
| HumanEval | ~93% | ~90% |
| Agentic (long-horizon) | Lead | Competitive |
| Context window | 200K | 1M |
| License | Proprietary | MIT (open weights) |
Opus 4.8 numbers are Anthropic's published figures; V4 Pro numbers from DeepSeek's May 1, 2026 preview. Independent re-benchmarks pending.
Pricing: Where DeepSeek Still Wins by an Order of Magnitude
The headline change Anthropic emphasised today is the cheaper fast mode, not a base price cut. Opus 4.8 standard pricing is still in line with Opus 4.7 — premium territory.
| Price (per 1M tokens) | Claude Opus 4.8 | DeepSeek V4 Pro |
|---|---|---|
| Input | ~$15.00 | $0.435 |
| Output | ~$75.00 | $0.87 |
| Effective multiple | 1× | ~17–86× cheaper |
Even after the new cheaper fast tier, a workload that costs $10,000/month on Opus 4.8 would cost roughly $120–600/month on DeepSeek V4 Pro. For agentic loops that burn millions of output tokens per day, that's the difference between an experiment and a P&L line item.
See the full breakdown in our V4 Pro 75% price cut analysis.
Which One Should You Pick in 2026?
Pick Claude Opus 4.8 when…
- You need the strongest agentic judgement available today.
- Your team lives inside Claude Code's new dynamic workflows.
- Compliance demands a US/UK-hosted closed-weights model.
Pick DeepSeek V4 Pro when…
- API cost is anywhere in your top-3 constraints.
- You need a 1M-token context window.
- You want MIT-licensed open weights for self-hosting.
- You're shipping high-volume chat, RAG, or batch inference.
Most teams will end up using both: Opus 4.8 for the hardest 5% of agentic problems, V4 Pro for the other 95%. The pricing gap is simply too large to ignore — and DeepSeek's open weights mean you also keep an exit ramp from any single vendor.
FAQ
When was Claude Opus 4.8 released?
May 28, 2026 — available today on claude.ai, in Claude Code, and via the Anthropic API at the same price as Opus 4.7.
Is Claude Opus 4.8 cheaper than DeepSeek V4 Pro?
No. V4 Pro is roughly 17–86× cheaper per token. Opus 4.8's pricing news is about the new fast mode being 3× cheaper than before, not a base price cut.
What are dynamic workflows in Claude Code?
A new feature that lets Opus 4.8 plan and execute very large-scale engineering problems autonomously, orchestrating its own sub-tasks.
Should I switch from DeepSeek V4 Pro to Opus 4.8?
For most production workloads, no — V4 Pro stays the best value. Switch (or add) Opus 4.8 only when you need its strongest-in-class agentic judgement.