Step 5 Preview vs DeepSeek V4 Pro
Official StepFun and DeepSeek API prices and LMArena scores, checked 2026-11-01. Add up to 2 more models to compare 4.
Step 5 Preview costs less; LMArena tasks are split evenly.
Step 5 Preview is 1.3x cheaper on input and 1.5x cheaper on output. Step 5 Preview scores 11 points higher for coding; DeepSeek V4 Pro scores 1 point higher for creative writing. Across 5 shared LMArena tasks, they split 1 to 1 (3 tied). A medium month (20M input + 5M output tokens) costs $33.50 on Step 5 Preview and $46.20 on DeepSeek V4 Pro.
- Pick Step 5 Preview if you want the lower bill ($33.50 vs $46.20 for a medium month) or you care most about coding (+11).
- Pick DeepSeek V4 Pro if you reuse long prompts (cached input $0.044 vs $0.10).
Medium: 20M input + 5M output tokens a month at list price (short-context rate, no caching or batch discounts).
Marked side is better: lower for prices, higher for context and scores. Bars are scaled to the larger of the two.
| Spec | Step 5 Preview | DeepSeek V4 Pro |
|---|---|---|
| Price per 1M tokens | ||
| Input | $1 | $1.32 |
| Cached input | $0.10 | $0.044 |
| Output | $2.70 | $3.96 |
| Blend 3:1 | $1.43 | $1.98 |
| Sample month, medium | $33.50 | $46.20 |
| Context | ||
| Context window | Not published | 1M tokens |
| Quality (LMArena, out of 100) | ||
| Coding | 11/100 #28 of 50 | 0/100 #38 of 50 |
| Chat | 0/100 #32 of 51 | 0/100 #26 of 51 |
| Creative writing | 0/100 #34 of 51 | 1/100 #19 of 51 |
| Roleplay and characters | 0/100 #38 of 51 | 0/100 #20 of 51 |
| Reasoning and math | 0/100 #28 of 51 | 0/100 #38 of 51 |
| Vision | 0/100 #23 of 33 | Not ranked |
| Long documents | Not ranked | Not ranked |
| Cheap bulk and agents | 21/100 #28 of 37 | Not ranked |
| Details | ||
| Provider | StepFun | DeepSeek |
| API model id | step-5-preview | deepseek-v4-pro |
| Notes | Preview. The input price includes writing new content to the cache. | Peak rate. Off-peak is half price. |
| Sources | ||
| Last checked | 2026-11-01 | 2026-11-01 |
| Official pricing | StepFun pricing | DeepSeek pricing |
Plans from data checked hourly against each vendor's official pricing page. Vendors state limits in different units, so we quote them as published. Team and enterprise plans are per seat.
API prices in USD per 1M tokens from each provider's official pricing page. Blend 3:1 = (3 x input + output) / 4. Sample month = stated input and output tokens at the list price, no caching. Quality: relative score out of 100 on each LMArena leaderboard (100 = top model we price on that board, n/a = not ranked there), data licensed CC BY 4.0. See how we pick.
Where each one ranks on LMArena
Relative score out of 100 on each leaderboard (100 = the top model we price there), with the rank among the models from our API price table that appear on that board.
- Coding (Code Arena): Step 5 Preview #28 of 50 (11/100), DeepSeek V4 Pro #38 of 50 (0/100)
- Chat (Text Arena): Step 5 Preview #32 of 51 (0/100), DeepSeek V4 Pro #26 of 51 (0/100)
- Creative writing (Text Arena creative writing): Step 5 Preview #34 of 51 (0/100), DeepSeek V4 Pro #19 of 51 (1/100)
- Roleplay and characters (Text Arena multi-turn): Step 5 Preview #38 of 51 (0/100), DeepSeek V4 Pro #20 of 51 (0/100)
- Reasoning and math (Text Arena math): Step 5 Preview #28 of 51 (0/100), DeepSeek V4 Pro #38 of 51 (0/100)
- Vision (Vision Arena): Step 5 Preview #23 of 33 (0/100), DeepSeek V4 Pro not ranked
- Long documents (Document Arena): Step 5 Preview not ranked, DeepSeek V4 Pro not ranked
- Cheap bulk and agents (Agent Arena): Step 5 Preview #28 of 37 (21/100), DeepSeek V4 Pro not ranked
Pricing details from the official pages
Step 5 Preview lists $1 input, $2.70 output, $0.10 cached input per 1M tokens on StepFun pricing. DeepSeek V4 Pro lists $1.32 input, $3.96 output, $0.044 cached input per 1M tokens on DeepSeek pricing. Prices checked 2026-11-01.
- Step 5 Preview: Preview. The input price includes writing new content to the cache.
- DeepSeek V4 Pro: Peak rate. Off-peak is half price.
Sample monthly bill at list price, no caching: Light (2M input + 0.5M output tokens): $3.35 vs $4.62. Medium (20M input + 5M output tokens): $33.50 vs $46.20. Heavy (200M input + 50M output tokens): $335 vs $462.
FAQ
Is Step 5 Preview cheaper than DeepSeek V4 Pro?
Step 5 Preview costs $1 input and $2.70 output per 1M tokens; DeepSeek V4 Pro costs $1.32 and $3.96. For a medium month (20M input + 5M output tokens) Step 5 Preview is cheaper: $33.50 vs $46.20.
Which is better for coding, Step 5 Preview or DeepSeek V4 Pro?
On LMArena's Code Arena Step 5 Preview scores 11/100 (#28 of 50) and DeepSeek V4 Pro 0/100 (#38 of 50), so Step 5 Preview is ahead by 11 points.
Which has the bigger context window?
Step 5 Preview: n/a. DeepSeek V4 Pro: 1M tokens.
How much do Step 5 Preview and DeepSeek V4 Pro cost per month?
Light (2M input + 0.5M output tokens): $3.35 vs $4.62. Medium (20M input + 5M output tokens): $33.50 vs $46.20. Heavy (200M input + 50M output tokens): $335 vs $462. Pick your own token counts in the plan vs API calculator.
Where do these numbers come from?
Prices come from each provider's official pricing page, checked daily. Quality is the relative score on the LMArena leaderboard dataset, licensed CC BY 4.0. Where a model is not ranked we show n/a.
Sources
Related
- Cheapest LLM APIs, compared
- StepFun API pricing, every model
- DeepSeek API pricing, every model
- DeepSeek V4 Pro vs GPT-6.1 Sol
- Qwen3.8 Max vs DeepSeek V4 Pro
- Mistral Large 4 vs DeepSeek V4 Pro
- DeepSeek V4 Pro vs Claude Opus 5.5
- Kimi K3.1 Pro vs DeepSeek V4 Pro
- GLM-5.3 vs DeepSeek V4 Pro
- DeepSeek V4 Pro vs Gemini 4 Argon