Side-by-side comparison of pricing, 12 benchmarks, and generation speed.
| Metric | DeepSeek R1 0528 (May '25) | Claude Opus 4.6 (Non-reasoning, High Effort) |
|---|---|---|
| Input ($/M tokens) | $1.35 | $5 |
| Output ($/M tokens) | $4.2 | $25 |
Verdict. Claude Opus 4.6 (Non-reasoning, High Effort) takes the aggregate benchmark matchup 6–5 across 11 categories. Real workloads usually care about a handful of specific tasks — see the per-benchmark table above.
Pricing. Both models sit in the mid-tier / premium bracket for output-token pricing. At 0.2× the per-million-token cost, DeepSeek R1 0528 (May '25) is meaningfully cheaper if your traffic is output-heavy (long completions, document generation, agent loops). DeepSeek R1 0528 (May '25) makes more sense when output volume is low and absolute reasoning quality justifies the premium.
Strengths. DeepSeek R1 0528 (May '25) is strongest on MATH-500 (98%), MMLU-Pro (85%), GPQA Diamond (81%). Claude Opus 4.6 (Non-reasoning, High Effort) leads on GPQA Diamond (84%), TerminalBench (48%), SciCode (46%).
Speed. Throughput is comparable — 0 tok/s vs 0 tok/s — so generation speed shouldn't drive your choice here. Look at the per-benchmark wins instead.
Provider. DeepSeek and Anthropic sell to overlapping but distinct developer audiences: DeepSeek tends to ship frontier reasoning models with premium positioning, while Anthropic often prices more aggressively. Your existing vendor relationships, billing, and SLA preferences may matter as much as the raw numbers above.
Workload cost. Workload scenarios (per million requests at 30M input + 15M output tokens): DeepSeek R1 0528 (May '25) costs $103.50 ($1242/year); Claude Opus 4.6 (Non-reasoning, High Effort) costs $525.00 ($6300/year). At a smaller 5M-input/2M-output scale (single-developer tool or prototype): DeepSeek R1 0528 (May '25) ≈ $15.15/run, Claude Opus 4.6 (Non-reasoning, High Effort) ≈ $75.00/run. At agent/realtime scale (200M input / 100M output per million requests): DeepSeek R1 0528 (May '25) ≈ $690/run, Claude Opus 4.6 (Non-reasoning, High Effort) ≈ $3500/run. DeepSeek R1 0528 (May '25) becomes more attractive at higher volume — the absolute per-token pricing difference compounds when you ship at scale.
Recommendation. Benchmarks are too close to call (1-win swing) but Claude Opus 4.6 (Non-reasoning, High Effort) is 6.0× more expensive per million output tokens. For most workloads the cheaper option (DeepSeek R1 0528 (May '25)) wins on cost-quality tradeoff. Only pay the premium for Claude Opus 4.6 (Non-reasoning, High Effort) if you have a measured lift on a task your product depends on.
Head-to-head deltas
Data from Artificial Analysis API — 12 benchmarks
DeepSeek R1 0528 (May '25) is cheaper overall. Its blended price (3:1 input/output ratio) is $2.06/M tokens vs $10.00/M for Claude Opus 4.6 (Non-reasoning, High Effort).
Claude Opus 4.6 (Non-reasoning, High Effort) wins 6 out of 12 benchmarks compared to 5 for DeepSeek R1 0528 (May '25). See the detailed benchmark chart above for per-category results.
Both models have comparable generation speeds.
Choose based on your priorities: DeepSeek R1 0528 (May '25) for lower cost, Claude Opus 4.6 (Non-reasoning, High Effort) for stronger benchmark performance, and both have comparable speed. For latency-sensitive apps, check the TTFT comparison above.