Compare / head-to-head
Claude Sonnet 5.5vs
DeepSeek V4 Flash
Claude Sonnet 5.5 leads 16 of 17 shared benchmarks. Both offer a 1M-token context window.
Benchmarks from cited public sources; pricing from official pages; status from official provider feeds.
Shared benchmarks
16 – 1
Claude Sonnet 5.5 leads
Cheaper per token
—
list price, input + output
Larger context
Tie
both 1M tokens
Providers
2 providers
Anthropic · DeepSeek
Claude Sonnet 5.5
Anthropic · released 2026-09-28
textvision
DeepSeek V4 Flash
DeepSeek · released 2026-07-31
text
- Context
- 1M
- Max out
- —
- Input /1M
- —
- Output /1M
- —
- Cached /1M
- —
- Scores
- 28 · 7 core
Quality
Benchmark matrix
17 shared · 12 only Claude Sonnet 5.5 · 11 only DeepSeek V4 Flash| Benchmark | Claude Sonnet 5.5 | DeepSeek V4 Flash | Δ | Edge |
|---|---|---|---|---|
| Reported by both · 17 | ||||
| GPQA Diamond | 95.6% ↗epoch run | 91% ↗epoch run | +4.6 pt | Claude Sonnet 5.5 |
| LMArena Elo | 1471 ↗ | 1436 ↗ | +35 | Claude Sonnet 5.5 |
| FrontierMath Tier 4 v2 (Epoch AI run) | 80.5% ↗epoch run | 24.4% ↗epoch run | +56.1 pt | Claude Sonnet 5.5 |
| FrontierMath Tiers 1-3 v2 (Epoch AI run) | 88.8% ↗epoch run | 57.5% ↗epoch run | +31.3 pt | Claude Sonnet 5.5 |
| LiveBench Agentic Coding (LiveBench) | 56.3% ↗ | 46.8% ↗ | +9.5 pt | Claude Sonnet 5.5 |
| LiveBench Coding (LiveBench) | 91.4% ↗ | 75% ↗ | +16.4 pt | Claude Sonnet 5.5 |
| LiveBench Data Analysis (LiveBench) | 78.6% ↗ | 79.3% ↗ | -0.7 pt | DeepSeek V4 Flash |
| LiveBench Instruction Following (LiveBench) | 70.5% ↗ | 65.5% ↗ | +5 pt | Claude Sonnet 5.5 |
| LiveBench Language (LiveBench) | 83.4% ↗ | 79.2% ↗ | +4.2 pt | Claude Sonnet 5.5 |
| LiveBench Mathematics (LiveBench) | 96.7% ↗ | 86.8% ↗ | +9.9 pt | Claude Sonnet 5.5 |
| LiveBench Reasoning (LiveBench) | 91.6% ↗ | 86.6% ↗ | +5 pt | Claude Sonnet 5.5 |
| LMArena WebDev (LMArena) | 1786.3 ↗ | 1581 ↗ | +205.3 | Claude Sonnet 5.5 |
| Mystery Game Puzzles (Epoch AI run) | 65% ↗epoch run | 34% ↗epoch run | +31 pt | Claude Sonnet 5.5 |
| OTIS Mock AIME 2024-2025 (Epoch AI run) | 100% ↗epoch run | 94.4% ↗epoch run | +5.6 pt | Claude Sonnet 5.5 |
| SimpleBench (SimpleBench) | 75.9% ↗ | 61.1% ↗ | +14.8 pt | Claude Sonnet 5.5 |
| SimpleQA Verified | 46.5% ↗epoch run | 33.6% ↗epoch run | +12.9 pt | Claude Sonnet 5.5 |
| Terminal-Bench 4.0 (Vals AI) | 64.1% ↗ | 18.7% ↗ | +45.4 pt | Claude Sonnet 5.5 |
| Only Claude Sonnet 5.5 reports · 12 | ||||
| APEX-Agents (Mercor) | 75.5% ↗ | not reported | — | — |
| Chartography (no tools) | 61.6% ↗ | not reported | — | — |
| CursorBench 4.0 | 55.5% ↗ | not reported | — | — |
| FrontierCode v1.1 (Main) | 46.2% ↗ | not reported | — | — |
| FrontierSWE V2 (Proximal Labs) | 61.9% ↗ | not reported | — | — |
| Furniture Assembly (Epoch AI run) | 75% ↗epoch run | not reported | — | — |
| Humanity's Last Exam (with tools) | 64.5% ↗ | not reported | — | — |
| LMArena Agent (LMArena) | 0.1252 ↗ | not reported | — | — |
| LMArena Vision (LMArena) | 1268.3 ↗ | not reported | — | — |
| OSWorld 2.1 (partial) | 80.1% ↗ | not reported | — | — |
| SAGE (Vals AI) | 51.8% ↗ | not reported | — | — |
| Terminal-Bench 4.0 | 70.6% ↗ | not reported | — | — |
| Only DeepSeek V4 Flash reports · 11 | ||||
| Agents' Last Exam | not reported | 25.2% ↗ | — | — |
| AutomationBench Public | not reported | 25.1% ↗ | — | — |
| Chess Puzzles (Epoch AI run) | not reported | 33% ↗epoch run | — | — |
| Cybergym | not reported | 76.7% ↗ | — | — |
| DeepSWE | not reported | 54.4% ↗ | — | — |
| DeepSWE v1.1 (Datacurve) | not reported | 53.3% ↗ | — | — |
| NL2Repo | not reported | 54.2% ↗ | — | — |
| Terminal Bench 2.1 | not reported | 82.7% ↗ | — | — |
| Toolathlon-Verified | not reported | 70.3% ↗ | — | — |
| Toolathlon-Verified (HKUST) | not reported | 70.7% ↗ | — | — |
| WeirdML (Håvard Tveit Ihle) | not reported | 63% ↗ | — | — |
Scores tagged "epoch" or "matharena" are independent runs, used only where the lab has not published its own; ⚠ marks rows MathArena flags as released after the competition. Higher is better on every row. Δ is Claude Sonnet 5.5 minus DeepSeek V4 Flash in the benchmark's own unit. "Not reported" means the lab has not published that figure; it is not a zero. ↗ opens the source.
Specs & pricing
Side by side
Official pricing pages and model cards| Spec | Claude Sonnet 5.5 | DeepSeek V4 Flash | Edge |
|---|---|---|---|
| Context window | 1M tokens | 1M tokens | Tie |
| Max output | 128K tokens | — | — |
| Input price / 1M | $2 ↗ | — | — |
| Output price / 1M | $10 ↗ | — | — |
| Cached input / 1M | $0.2 ↗ | — | — |
| Input + output / 1M Lower is cheaper. List prices; batch, tool and regional fees excluded. | $12 | — | — |
| Modalities | text · vision | text | Claude Sonnet 5.5 |
| Released | 2026-09-28 | 2026-07-31 | — |
| Cited benchmark scores | 29 | 28 | — |
Reliability
Provider status
Live from /statusMore matchups
Claude Sonnet 5.5 vs …
Models sharing the most benchmarksMore matchups
DeepSeek V4 Flash vs …
Models sharing the most benchmarksBuilt by Respan
Which one wins on your data?
Public benchmarks are a starting point. Run Claude Sonnet 5.5 and DeepSeek V4 Flash on your own prompts with Respan evals, or route to either through one gateway key with automatic failover.