Last updated:
DeepSeek V4.1 Flash vs Gemini 3.8 Flash vs Claude Haiku 5.5
Three fast, cheap models people compare for agents and high-volume API work. They are priced very differently and their vendors publish different benchmarks, so this page lines up what can actually be compared.
Specs and prices side by side
| DeepSeek V4.1 Flash | Gemini 3.8 Flash | Claude Haiku 5.5 | |
|---|---|---|---|
| Input / output per 1M | $0.15 / $0.60 off-peak, $0.30 / $1.20 peak | $0.75 / $3.75 (launch price to Dec 31, 2026; $1.50 / $7.50 after) | $0.10 / $0.50 up to 100K; $0.50 / $2.50 above |
| Cached input per 1M | $0.003 / $0.006 | $0.075 | $0.01 ($0.05 above 100K) |
| Context / max output | 1M / 384K | 1,048,576 / 65,536 | 1M / 128K |
| Inputs | Text, images | Text, images, video, audio | Text, images |
| Open weights | Yes (MIT) | No | No |
| Released | Sept 10, 2026 (1017 update Oct 17) | Sept 2, 2026 | Oct 7, 2026 |
Source: DeepSeek Models & Pricing
Source: Gemini API pricing
Source: Gemini 3.8 Flash model page
Source: Anthropic pricing doc
What the same workload costs
A: 1,000 calls of 20K input tokens (half cached) and 2K output. B: 100 long-context calls of 200K input tokens (90% cached) and 4K output. List prices, cache-write and storage fees ignored.
| Model | Workload A | Workload B |
|---|---|---|
| DeepSeek V4.1 Flash, off-peak | $2.73 | $0.59 |
| DeepSeek V4.1 Flash, peak | $5.46 | $1.19 |
| Claude Haiku 5.5 | $2.10 | $2.90 |
| Gemini 3.8 Flash (launch price) | $15.75 | $4.35 |
Each model counts tokens with its own tokenizer and thinks for a different number of tokens, so the real bill per task can differ from these list-price examples.
Source: DeepSeek Models & Pricing
Source: Anthropic pricing doc
Source: Gemini API pricing
Benchmarks: what can be compared
The vendors mostly publish different tests. These rows are the overlap; vendor numbers come from each company’s own runs, so small gaps mean little.
| Test | DeepSeek V4.1 Flash | Gemini 3.8 Flash | Claude Haiku 5.5 |
|---|---|---|---|
| Artificial Analysis Intelligence Index | 39 | 41 | 43 |
| AA output speed (tokens/s) | 175.9 | 131.5 | 235.8 |
| AA cost per index task | $0.27 (peak price) | $1.24 | $0.21 |
| Terminal-Bench 4.0 (vendor) | 42.3 (1017) | Not published | 39.2 (max effort) |
| Humanity's Last Exam, no tools (vendor) | 36.8 | Not published | 45.9 |
| Humanity's Last Exam, with tools (vendor) | 63.9 | Not published | 57.4 |
Artificial Analysis figures read on 2026-10-13 (Gemini, Haiku) and 2026-10-17 (DeepSeek, September release at Max effort).
Source: Artificial Analysis: DeepSeek V4.1 Flash
Source: Artificial Analysis: Gemini 3.8 Flash
Source: Artificial Analysis: Claude Haiku 5.5
Source: DeepSeek API change log
Source: Hugging Face model card
Source: Anthropic: Claude Haiku 5.5
Which one to pick
- Pick DeepSeek V4.1 Flash. Long-context agents, heavy prompt caching, overnight or weekend batch jobs, and anyone who needs open weights or self-hosting.
- Pick Claude Haiku 5.5. Short, high-volume calls under 100K tokens, the fastest output of the three, and teams already on the Claude API or Claude Code.
- Pick Gemini 3.8 Flash. Audio and video input, Google Search grounding, and prototyping on the free tier.
DeepSeek V4.1 Flash guide · Gemini 3.8 Flash guide · Claude Haiku 5.5 guide
FAQ
Is DeepSeek V4.1 Flash cheaper than Claude Haiku 5.5?
Not for short prompts: Haiku 5.5 lists $0.10 / $0.50 against DeepSeek’s $0.15 / $0.60 off-peak. Above 100K-token prompts Haiku rises to $0.50 / $2.50, and DeepSeek’s cache hits are cheaper, so long-context work usually costs less on DeepSeek.
Is DeepSeek V4.1 Flash cheaper than Gemini 3.8 Flash?
Yes, at list price. Gemini 3.8 Flash costs $0.75 / $3.75 per 1M tokens at its launch price, five times DeepSeek’s off-peak input price and over six times its output price.
Which is smartest?
They are close. Artificial Analysis scores Haiku 5.5 at 43, Gemini 3.8 Flash at 41 and DeepSeek V4.1 Flash at 39; DeepSeek’s 1017 update came after that test.
Which one has open weights?
Only DeepSeek V4.1 Flash, under the MIT licence.
Independent guide, not affiliated with DeepSeek, Google or Anthropic. Prices and model versions change often; check the linked official pages before you build on them.