The open‑source Chinese model DeepSeek V3 climbs to #4 in our creative writing leaderboard, delivering lean imaginative prose at a bargain‑basement price.
DeepSeek V3 lands fourth (Δ+3), pairing vivid Chinese‑English creativity with the cheapest tokens in the top tier.
Ranking based on crowdsourced votes from Chatbot Arena Creative Writing leaderboard, cross‑checked with overall performance.
64000 tokens
Yes
Moderate safety filters; more permissive than most closed models
2024-07
Style Comment:
Write English lyrics for a mid-tempo pop song (≈ 100 BPM) in ABAB CB structure (two verses, pre-chorus, chorus, bridge, final chorus).
Theme: personal resilience after social rejection.
Requirements
• Chorus must contain the hook line “I bend, I never break.”
• Bridge should introduce a fresh metaphor and shift the rhyme scheme.
• Avoid explicit profanity; keep radio-friendly.
• Deliver as: V1 / V2 / Pre-Ch / Ch / Bridge / Ch. Only the lyrics, no chords.
Style Comment:
Compose a Shakespearean sonnet (14 lines, iambic pentameter, abab cdcd efef gg) on the tension between digital memory and human forgetting. Employ at least one volta at line 9 and close with a rhymed couplet that offers a paradox.
Style Comment:
Write a 1 000-word SEO-optimized blog article aimed at first-time home coffee-roasters.
Requirements
Style Comment:
Write high-conversion landing-page copy (≈ 300 words) for a new line of lightweight, fully recyclable running shoes named Zephyr-Lite™.
Audience → urban runners, ages 20-40, eco-conscious but style-driven.
Must include
• A punchy headline (≤ 8 words)
• Sub-headline elaborating the core benefit (≤ 18 words)
• 3 scannable benefit bullets (2-3 lines each)
• One short customer testimonial (first name, city)
• A single CTA button label (≤ 4 words)
Tone: energetic, conversational, modern.
Style Comment:
Write an 800-word argumentative essay assessing whether universal basic income (UBI) would stimulate entrepreneurship in developed economies.
Requirements
• Present a clear thesis in the introduction.
• Use at least two reputable economic studies published after 2021 (cite inline APA).
• Address one major counterargument.
• Conclude with a policy recommendation.
• Formal academic tone, but readable. Return only the essay text.
Style Comment:
Generate a structured mini-literature review (≈ 1 000 words) on the latest (2022-2025) CRISPR-based therapies for sickle-cell disease.
Deliver as:
Cite sources in IEEE style; include DOI where available. Output only the review.
DeepSeek V3 writes with concise, image‑rich Chinese English, favoring stoic narrative arcs and tight two‑beat lyricism; technical prose is factual but sometimes light on detail.
Downloadable checkpoints under permissive MIT license enable full local fine‑tuning.
Input $0.27/M and output $1.10/M with 50 % off‑peak discount plus prompt cache.
Native support in vLLM, LMDeploy and TensorRT‑LLM unlocks high‑throughput serving.
DeepSeek’s API is the market’s bargain: just $0.27 per million input tokens and $1.10 per million output tokens, with 50 % off during off‑peak hours.
Input (cache miss): $0.27/M; Cache‑hit: $0.07/M; Output: $1.10/M; Off‑peak 50 % discount 16 :30–00 :30 UTC.
If DeepSeek V3 doesn’t fit, consider these competitors:
Best for: Premium creativity with tool calling
OpenAI’s flagship with multimodal reasoning and 128 K context.
Read ReviewBest for: Sensitive narrative or policy drafts
Empathetic long‑form writing and moral poetry.
Best for: Research‑heavy content
Google’s 1 M‑token context powerhouse with real‑time web access.
Read ReviewYes. The weights and code are released on GitHub under the MIT license.
Up to 64 K tokens per request.
July 2024.
Input: $0.27/M; Output: $1.10/M; Cache hit: $0.07/M; Off‑peak 50 % discount.
No, DeepSeek V3 is text‑only at the moment.
Overall, DeepSeek V3 offers unmatched price‑performance for creative writing, backed by open weights and a permissive license. Teams needing ultra‑long context or blazing speed may prefer pricier alternatives.