Qwen3.8-Max vs Qwen3.8-Flash: What the Tier Buys

All figures below are from the China AI Hub database, last verified 2026-09-20 (Flash last verified 2026-09-27). Where a field is not publicly disclosed, we say so rather than estimating.

At a glance

Dimension	Qwen3.8-Max	Qwen3.8-Flash
Status (database)	active	active
Context window	1,048,576 tokens	1,048,576 tokens
Maximum output	131,072 tokens	131,072 tokens
Input price (per 1M)	$2.00 (Singapore)	$0.15 (Singapore)
Output price (per 1M)	$6.00 (Singapore)	$0.47 (Singapore)
Open weights	No (proprietary)	No (proprietary)
API available	Yes	Yes
Reasoning	Yes	Yes
Coding	Yes	Not listed
Vision / Video	Yes / Yes	Yes / Yes
Tool calling	Yes	Not listed
Structured output	Yes	Not listed
Benchmark records in DB	5	0

Pricing

Qwen3.8-Max lists $2.00 input / $6.00 output per 1M tokens; Qwen3.8-Flash lists $0.15 input / $0.47 output — a fraction of the flagship price on both axes (roughly 1/13 the input and 1/13 the output). Both are Singapore-region list prices; the Beijing/Global regions list lower rates ($1.65/$4.951 for Max, $0.113/$0.382 for Flash) — a roughly 14x gap on both axes that holds across regions. The price spread between Alibaba’s two tiers is one of the widest same-vendor gaps in the database, and context caching is available on both to reduce effective cost on re-read-heavy workloads.

Context and output

Identical listed context: 1,048,576 tokens input (991,808 max input documented for Max), 131,072 maximum output. On this axis the two are equivalent per the database.

Architecture and parameters

Qwen3.8-Max is a 2.4T-parameter MoE with 95B activated (512 experts, 10 routed + 1 shared per token, 92 layers) using a Gated DeltaNet + Gated Attention hybrid. Alibaba does not publicly disclose architecture or parameter counts for the API-only Qwen3.8-Flash — the database records that absence rather than estimating it. This is the standard split across the Qwen3.8 family: the flagship carries a published architecture card, while the flash tier is documented only by capability and price.

Coding

This is the clearest recorded split. Qwen3.8-Max lists coding and publishes vendor-reported Terminal-Bench 2.1 (86.6) and SWE-bench Pro (67.7). Qwen3.8-Flash does not list coding in its documented capability set (its frontmatter records only reasoning, vision and video). For coding workloads the database records a difference in listed capability and a total absence of benchmark evidence for the flash model.

Vision

Both list vision and video input. On multimodal input the two are equivalent per the database.

Agent capabilities

Qwen3.8-Max lists tool calling, structured output, function calling and web search — a fuller agent substrate. Qwen3.8-Flash lists none of these in its recorded capability set. The Max model is documented with prefix completion and context caching in addition to the tool surface, which is why it serves as Alibaba’s agent substrate; the Flash tier is documented only as a low-cost multimodal endpoint. For tool-driven agent pipelines, only the flagship lists the relevant surface.

License and openness

Both are proprietary, closed weight, no self-hosting. This axis does not differentiate them. Both are documented as OpenAI- and Anthropic-API-protocol compatible (Max per Model Studio; Flash per Alibaba’s own statement).

Deployment

Both are served through Alibaba Cloud Model Studio across the same six regions (Beijing, Singapore, Hong Kong, Frankfurt, US-Virginia, Tokyo). Batch inference is supported on Max and not on Flash. Both are documented as OpenAI- and Anthropic-API-protocol compatible (Max per Model Studio; Flash per Alibaba’s own statement), which matters for drop-in migration of existing tooling. Deployment is otherwise equivalent.

Benchmarks

The database holds 5 benchmark records for Qwen3.8-Max — Terminal-Bench 2.1 (86.6), SWE-bench Pro (67.7), GPQA Diamond (92.6), HLE (43.6, 56.2 with tools) and MRCR v2 256K (92.9) — and 0 for Qwen3.8-Flash. There is no benchmark evidence recorded for the flash model. The Max scores are vendor-reported from its model card; the 0902 snapshot (dated 2026-09-02) upgraded coding and vision at an unchanged price.

Benchmark comparability is limited: Qwen3.8-Flash has no benchmark records, so no head-to-head score comparison is possible; Qwen3.8-Max’s scores are vendor-reported from its model card.

Why each difference matters

China AI Hub analysis indicates the following per-dimension implications, drawn from the listed facts above.

• Pricing: Qwen3.8-Flash has an advantage at a fraction of Max’s price on both axes; the output-token gap compounds on reasoning workloads.
• Context: identical 1M-token windows — this axis does not differentiate.
• Coding: Qwen3.8-Max is more relevant for coding workloads — it lists coding and coding benchmarks where Flash lists neither.
• Vision / media: both list vision and video — this axis does not differentiate.
• Agent: Qwen3.8-Max is more relevant for tool-driven agent pipelines — it lists tool calling, structured output and web search where Flash lists none.
• License: both proprietary closed weight — this axis does not differentiate.
• Deployment: equivalent six-region footprint; Max supports batch inference where Flash does not.
• Benchmark: Max records 5 and Flash records 0, so the evidence base is not comparable.

Trade-off summary

• Price: Qwen3.8-Flash lists a fraction of Max’s input and output price.
• Coding and tools: Qwen3.8-Max lists coding, tool calling and structured output; Flash lists none of these.
• Vision and context: identical (vision/video, 1M context).
• Both proprietary closed weight, same six regions.

Choose by workload: high-volume, multimodal and low-cost work favors Qwen3.8-Flash’s price; coding, tool-driven and benchmark-backed workloads favor Qwen3.8-Max’s listed capability set. Verify current prices on the official pages before committing.

Decision context

China AI Hub analysis indicates the following decision-context implications, drawn from the listed facts above.

For API developers. Qwen3.8-Flash lists ~1/13 the input and output price, so cost favors the flash tier for high-volume work. Qwen3.8-Max lists coding, tool calling and structured output, which Flash does not; both are OpenAI/Anthropic-protocol compatible. Latency is not publicly documented for either model on this page.

For self-hosting. Neither is self-hostable — both are proprietary closed weight. This axis is a wash.

For coding agents. Qwen3.8-Max lists coding, tool calling and structured output, which is more relevant for coding-agent pipelines; Qwen3.8-Flash lists none of these and has no benchmark records.

For enterprise. Deployment is equivalent (same six regions), but the capability split matters: Max lists coding and a fuller tool surface, while Flash is the budget tier. Region, SLA, data-residency and compliance terms are not publicly documented — confirm with the vendor.

What is uncertain

• Qwen3.8-Flash has no benchmark records and no documented coding capability — an absence of documentation, not a verified lack of capability.
• Qwen3.8-Flash’s architecture and parameter counts are not publicly disclosed.
• No third-party benchmark evidence is recorded for either model.

Sources

• Alibaba Cloud Model Studio — Qwen3.8-Max
• Alibaba Cloud Model Studio — Qwen3.8-Flash
• Alibaba Cloud Model Studio — Model pricing

Labels used above: Official fact (prices, context windows, capabilities and regions from primary provider sources), Vendor-reported claim (benchmark scores), and China AI Hub analysis (the “why each difference matters” reasoning, introduced as analysis). No third-party benchmark evidence is currently recorded for these models.