← ModelRumors.ai buyer's guide
HEAD TO HEAD · UPDATED NOV 6, 2026
DeepSeek V4-Pro vs Grok 4.7
DeepSeek against xAI, on sourced numbers only. Green marks the higher (or cheaper) value on each row; the text below only calls a lead when the gap is bigger than measurement noise.
| DeepSeek V4-Pro | Grok 4.7 | |
|---|---|---|
| Arena Elo | 1465 source ↗ | 1441 source ↗ |
| GPQA Diamond | 92.4% source ↗ | not published |
| $ per 1M in / out | $0.66 / $1.98 source ↗ | $2 / $6 source ↗ |
| Blended $ per 1M (3:1) | $0.99 source ↗ | $3.00 source ↗ |
| Context window | 1M source ↗ | 500K source ↗ |
| Output speed | not published | 72 tok/s source ↗ |
| Released | Aug 13, 2026 source ↗ | Sep 21, 2026 source ↗ |
| Open weights | Yes source ↗ | No source ↗ |
| Reasoning effort tiers | low · high · max (default: high) | low · medium · high · xhigh (default: high) |
| API effort control | thinking={"type": "enabled" | "disabled"}, reasoning_effort="low" | "high" | "max" (medium and xhigh map to high) source ↗ | reasoning_effort="low" | "medium" | "high" | "xhigh" source ↗ |
| Additional benchmarks | SWE-bench Verified: 80.6% (April V4 preview card, Max) source ↗ | DeepSWE v1.1: 71.0% source ↗ CursorBench 4.0: 46.3% source ↗ AA Intelligence Index v5.0.1: 35 (xhigh) source ↗ |
| Best for (workload) | Higher GPQA Diamond than V4.1-Flash (92.4% vs 90.9%, DeepSeek); MIT open weights. source ↗ | xAI's flagship. Reasoning is always on and cannot be turned off; 500K context; 71.0% on DeepSWE v1.1 (xAI). source ↗ |
Arena Elo from the arena.ai text leaderboard as of Oct 27, 2026 source ↗. Speed measured by Artificial Analysis and varies by provider.
Which one should I use?
For high-volume work where cost matters most, DeepSeek V4-Pro is about 3.0 times cheaper per token. For overall answer quality as rated by people in blind tests, DeepSeek V4-Pro leads clearly (1465 vs 1441 Arena Elo). If you need to self-host or fine-tune, only DeepSeek V4-Pro has open weights. On release timing, DeepSeek is at day 20 of a typical ~32-day gap (MID-CYCLE), and xAI is at day 46 of a typical ~46-day gap (PAST TYPICAL GAP).
Key differences from the sourced numbers
- DeepSeek V4-Pro is the cheaper option: $0.99 vs $3.00 per 1M tokens at a 3:1 input-to-output mix (about 3.0 times less).
- DeepSeek V4-Pro scores 24 Elo points higher (1465 vs 1441) on the Arena text leaderboard, where people vote on blind side-by-side answers.
- DeepSeek V4-Pro pricing note: Off-peak price (cache hit $0.022 / 1M). Peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC, weekdays): $1.32 / $3.96 (cache hit $0.044 / 1M). API continues until V4.1-Pro launches.
- Grok 4.7 pricing note: Prompts under 200K tokens; $4 / $12 above 200K.
Frequently asked questions
Is DeepSeek V4-Pro or Grok 4.7 better?
DeepSeek V4-Pro is the cheaper option: $0.99 vs $3.00 per 1M tokens at a 3:1 input-to-output mix (about 3.0 times less). DeepSeek V4-Pro scores 24 Elo points higher (1465 vs 1441) on the Arena text leaderboard, where people vote on blind side-by-side answers. DeepSeek V4-Pro pricing note: Off-peak price (cache hit $0.022 / 1M). Peak hours (01:00 to 04:00 and 06:00 to 10:00 UTC, weekdays): $1.32 / $3.96 (cache hit $0.044 / 1M). API continues until V4.1-Pro launches. Grok 4.7 pricing note: Prompts under 200K tokens; $4 / $12 above 200K.
Which one should I use: DeepSeek V4-Pro or Grok 4.7?
For high-volume work where cost matters most, DeepSeek V4-Pro is about 3.0 times cheaper per token. For overall answer quality as rated by people in blind tests, DeepSeek V4-Pro leads clearly (1465 vs 1441 Arena Elo). If you need to self-host or fine-tune, only DeepSeek V4-Pro has open weights. On release timing, DeepSeek is at day 20 of a typical ~32-day gap (MID-CYCLE), and xAI is at day 46 of a typical ~46-day gap (PAST TYPICAL GAP).
How much does DeepSeek V4-Pro cost compared with Grok 4.7?
DeepSeek V4-Pro lists at $0.66 / $1.98 per 1M input/output tokens ($0.99/1M blended at 3:1). Grok 4.7 lists at $2 / $6 per 1M input/output tokens ($3.00/1M blended at 3:1).
Facts checked Nov 5, 2026 against official sources. Every number links to its source. Spotted an error? Email [email protected].