Qwen3.8-Flash Review: $0.15/M Input, 67% Cheaper Than DeepSeek

📌 Key Takeaways

Launch then price cut: Qwen3.8-Flash launched on August 26 at $0.16/$0.47 per million tokens; Alibaba cut the China-side price on August 27 to ¥0.8 (input) and ¥2.7 (output), with cache hits at ¥0.1/M
Architecture breakthrough: 125B-parameter MoE model activating only 6B per token, plus a 51B N-gram embedding layer — training cost is ~1/9 of Qwen3.7-Plus (89% reduction)
Competitive performance: Scores 62.5 on SWE-bench Pro (agentic coding), leads in 8 of 14 benchmarks, matching DeepSeek-V4-Flash and Claude Opus 4.6 on real-world tasks
Mega context: Natively supports 262K tokens, extends to 1M with YaRN. Multimodal: text, image, and video input
Cache economics: Cache hit at $0.016/M tokens — ideal for agents, RAG, and long conversations

👉 Looking for cloud servers? Alibaba Cloud Up to 90% Off | Tencent Cloud Hot Deals | All Deals

💰 Latest Pricing (Effective August 27, 2026)

Pricing via Alibaba Cloud Model Studio (international) and Bailian platform (China), per million tokens.

International Pricing (USD)

Billing Item	Price (per 1M tokens)
Input	$0.15
Output	$0.47
Cache hit	$0.016

China Pricing (RMB, after Aug 27 cut)

Billing Item	Launch Price (Aug 26)	Current Price (Aug 27+)	Change
Input	¥1.0	¥0.8	↓20%
Output	¥3.0	¥2.7	↓10%
Cache hit	—	¥0.1	—

💡 Real-world cost example: An AI customer service app handling 1M requests/day (avg 2K input + 500 output tokens):

Qwen3.8-Flash: ~$415/day total

DeepSeek V4 Flash (peak): ~$1,260/day

Savings: ~$300,000+ per year

⚔️ Competitive Pricing Comparison

August 2026 pricing for major LLM APIs, USD per million tokens.

Model	Input	Output	Active Params	Context
Qwen3.8-Flash NEW	$0.15	$0.47	6B (MoE)	1M
Qwen3.5-Flash (≤128K)	$0.11	$0.67	—	1M
Qwen3.7-Flash	$0.034	$0.132	—	256K
Qwen3.8-Max	$2.00	$6.00	—	1M
DeepSeek V4 Flash (peak)	$0.42	$1.26	13B	128K
DeepSeek V4 Flash (off-peak 50%)	$0.21	$0.63	13B	128K
DeepSeek V4 Pro (peak)	$1.26	$3.78	—	128K
GLM-5.3-Flash	~$0.15+	~$0.47+	18B (MoE)	—

Key findings:
Qwen3.8-Flash output is 63% cheaper than DeepSeek V4 Flash peak ($0.47 vs $1.26)
Even against DeepSeek's off-peak pricing, Qwen3.8-Flash remains 25% cheaper on output
API price is 1/10 of GLM-5.3, and up to 1/20 during promotional periods
Qwen3.7-Flash is cheaper but lacks the 1M context and multimodal capabilities of 3.8

🏗️ Architecture Deep Dive

Qwen3.8-Flash (codenamed Qwen3.8-Flash-Next during development) is an early preview of the Qwen4 architecture, with four core innovations:

1. Mixture-of-Experts (MoE)

125B main parameters with only 6B activated per token (10 routed experts + 1 shared expert out of 512)
Additional 51B N-gram embedding layer stored in system RAM (not GPU), acting as a massive "phrase dictionary"
4B multi-token prediction parameters bring total footprint to ~180B

2. QSA Sparse Attention + GDN Hybrid

Qwen Sparse Attention (QSA) works alongside GDN: GDN compresses historical context while QSA selects key information. In high-cache-hit 1M-token scenarios, this delivers 7.6x faster prefill and 4.9x faster decode.

3. Gated Residual Mechanism

Splits the traditional single residual pathway into 4 parallel branches, dynamically gating information flow for better cross-layer communication and training stability.

4. Muon Optimizer

Uses a refined Muon + AdamW hybrid strategy with refitted scaling laws, eliminating batch warm-up and significantly improving convergence efficiency and training throughput.

Result: Training costs are approximately 1/9 of Qwen3.7-Plus (397B params, 17B activated) — an 89% reduction — while delivering superior coding and office task performance.

📊 Benchmark Performance

Per Qwen's official technical report, Qwen3.8-Flash excels across multiple evaluation suites:

Benchmark	Domain	Performance
SWE-bench Pro	Agentic coding	62.5, leading peers
CoWorkBench	Long-horizon office tasks	Beats DeepSeek V4 Flash
Toolathlon Verified	Real-world tool use	Matches Claude Opus 4.6
MathVision	Visual math reasoning	Strong multimodal
AndroidWorld	Mobile agent tasks	Embodied intelligence
ERQA	Embodied reasoning	Multimodal understanding

Across 14 evaluations, the base model (6B activated) achieved the best results in 8. The fine-tuned version shows even stronger performance in coding, agents, and multimodal tasks.

🎯 Best Use Cases

Qwen3.8-Flash is ideal for:
AI coding assistants: 62.5 on SWE-bench Pro, repository-level understanding with 1M context
AI agents / tool use: Native parallel tool calls, extremely low cost for long-horizon tasks
Customer service / chatbots: $0.016/M cache hit makes multi-turn conversations nearly free
Document analysis / RAG: 1M tokens processes hundreds of pages, cache speeds up 8x
Multimodal apps: Text + image + video input, visual math and chart analysis
High-volume batch processing: 5,000 RPM / 5M TPM rate limits for enterprise scale

🚀 How to Get Started

Option 1: Alibaba Cloud Model Studio (International)

Access Qwen3.8-Flash via Model Studio with OpenAI-compatible API. New users get free credit. The model serves on QwenCloud with 1M context by default.

Claim $200 Free Credit →

Option 2: Bailian Platform (China)

Available on Alibaba Cloud's Bailian platform with new user free token quota. First to receive the latest Qwen releases.

Bailian Console →

Option 3: Open Weights (Self-hosted)

The open-weight version, Qwen3.8-Flash-Next, is available on Hugging Face and ModelScope for download, fine-tuning, and local deployment. The production version with built-in tools and 1M default context is served via QwenCloud API.

New user offers:
Alibaba Cloud international: up to $200 free trial credit
Bailian platform: free token quota for new users (90 days)
ECS cloud servers: from ~$4.50/month (2 vCPU, 2GB RAM)

❓ FAQ

Qwen3.8-Flash vs Qwen3.8-27B — what's the difference?

Qwen3.8-Flash is a MoE model (125B total / 6B active) optimized for cost efficiency and high throughput. Qwen3.8-27B is a dense vision-language model (all 27B active) for deep reasoning at $0.42/$3.08 per million tokens. Choose Flash for scale, 27B for depth.

How does the cache hit pricing work?

When a request hits the context cache (e.g., repeated system prompts, long document prefixes), cached input tokens are billed at $0.016/M instead of $0.15/M — a 90% discount. This dramatically reduces costs for agents, RAG, and multi-turn conversations.

What input modalities are supported?

Qwen3.8-Flash supports text and image input natively (input_modalities: ["text", "image"]), with video analysis in the production version. It handles chart analysis, document OCR, and visual math.

What are the rate limits?

5,000 RPM (requests per minute) and 5,000,000 TPM (tokens per minute), suitable for enterprise-grade high-concurrency deployments.

Start Building with Qwen3.8-Flash Today

$200 free credit + ECS from $4.50/month + open-weight models available

Claim $200 Free Credit → View ECS Plans

Qwen3.8-Flash: Full Review