Claude Sonnet 5 vs DeepSeek V4
Pricing, benchmarks, and use case comparison
Newer versions are out: Claude Sonnet 5 has been replaced by Claude Sonnet 5.5.
Verdict
Our pick: Claude Sonnet 5Pick Claude Sonnet 5 for a managed, multimodal coder with strong computer use at $2/$10; pick DeepSeek V4 for MIT-licensed open weights, self-hosting, and top open-model reasoning at a lower API price.
Specs comparison
| Claude Sonnet 5 | DeepSeek V4 | |
|---|---|---|
| Provider | Anthropic | DeepSeek |
| Type | Closed source | Open source |
| Context window | 1M | 1M tokens |
| Input / 1M tokens | $2.00 | ✓$1.32 |
| Output / 1M tokens | $10.00 | $3.96 |
| Release date | 2026-06 | 2026-08 |
Benchmarks
| Benchmark | Claude Sonnet 5 | DeepSeek V4 |
|---|---|---|
| SWE-bench Verified | ~85.2% | - |
| Terminal-Bench 2.1 | 80.4% | - |
| SWE-bench Pro | 63.2% | - |
| OSWorld-Verified | 81.2% | - |
| SWE-bench Verified, April preview checkpoint | - | 80.6% |
| Math / STEM / Coding (open-model comparison) | - | Best among open models (per DeepSeek) |
| Terminal-Bench 4.0 | - | 12.4% (max effort) |
| DeepSWE v1.1 | - | 62.7% (max effort) |
| GPQA Diamond | - | 92.4% (max effort) |
Scores sourced from official provider release posts and independent benchmark aggregators.
Sonnet 5's successor is Claude Sonnet 5.5 at the same $2/$10 price, while DeepSeek V4 has no Pro successor yet: the API still serves the August V4-Pro-0813 release, and DeepSeek's newest model is the cheaper V4.1 Flash.
Capability and benchmarks
Both are strong agentic coders. Sonnet 5 reports ~85.2% SWE-bench Verified, 63.2% SWE-bench Pro, 80.4% Terminal-Bench 2.1, and 81.2% OSWorld-Verified. DeepSeek reported 80.6% SWE-bench Verified for its April V4-Pro preview and 92.4% GPQA Diamond for the August release, and says V4-Pro leads open models on math and STEM (math 93 on our index). Their coding scores on our index are close (Sonnet 92 vs DeepSeek 93), but Sonnet 5 adds image input (multimodal 85) and computer use, while DeepSeek V4 Pro is text-only (multimodal 10).
Price, context, and hosting
This is the core trade-off. DeepSeek V4 is open weight under MIT, free to self-host, with a 1M-token context and a hosted API that costs less per token than Sonnet 5, especially on output. Sonnet 5 is a managed model at $2/$10 (now its permanent price) with a 1M context, served on the Claude API, AWS, Google Cloud and Microsoft Foundry. If data control or per-token cost dominates, DeepSeek wins; if you want a turnkey API on the major clouds plus image input, Sonnet wins. The preview label that applied in July is gone: DeepSeek describes V4-Pro-0813 as the official release that supersedes the preview.
Which to pick
- Pick Sonnet 5 for managed agentic coding, image input, computer use, and availability across major clouds.
- Pick DeepSeek V4 for self-hosting or data sovereignty, top open-model reasoning and math, and a lower cost per token.
Which should you choose?
Choose Claude Sonnet 5 if...
- →You have a pipeline tuned to Sonnet 5 and want stable behavior while you test Sonnet 5.5
- →You turn thinking off with thinking type disabled or use forced tool_choice, both of which Sonnet 5.5 rejects
- →You need a 1M context window at $2/$10 with the Sonnet 5 request surface
Choose DeepSeek V4 if...
- →You need a frontier-class open model you can self-host for data control
- →Your workload involves very long documents, codebases, or agent trajectories (up to 1M tokens)
- →You want DeepSeek's largest model for knowledge-heavy text tasks
- →You need to fine-tune or customize a strong base model