Qwen 4 Max

What it is

Qwen 4 Max is Alibaba's flagship, and the note on this page calls it the strongest general open-weights option here, especially in multilingual work and coding. The Qwen family is the most prolific line in open models, and the Max tier is where Alibaba puts the general capability — broad coverage rather than one spike.

Whether this specific tier ships weights, and under what terms, should be verified on Alibaba's channels — the Qwen line has varied between open releases and API-only flagships across generations.

How the composite is built

Overall 85 weights Coding 24%, Terminal 20%, Reasoning 20%, Tool use 15%, Context 13%, Speed 8%. Qwen 4 Max runs 76 to 88 across the six axes — even by the standards of the upper middle of the board, with no axis below 76 and nothing above 88.

Coding 88

Level with Terra's 88 and eight behind Opus 5's 96. Strong general code work; Qwen 4 Coder's 90 is the specialist answer in the same family.

Terminal 82

One behind Llama 5's 83 and six behind Nemo 3 Ultra's 88. Capable, with supervision.

Reasoning 87

Its joint-strongest axis with Context: two behind Terra's 89 and six behind Gemini 3.6 Pro's 93. Good planning for the price class.

Tool use 82

One behind DeepSeek V4's 83 and eight behind Sonnet 5's 90. Structured calling works; long chains need validation.

Context 87

Behind the long-context specialists — Kimi k3's 92 and Gemini 3.6 Pro's 98 — but comfortably mid-frontier.

Speed 76

The bottom of its profile, level with Yi-2 Large's 76 band. Interactive use will feel like the upper-middle tier it is.

Where it fits

Multilingual products and general engineering work where one model has to cover code, analysis and writing without a specialist per task. For teams building in or for Chinese and other non-English markets, the Qwen line's language coverage is a durable reason to choose it over equally scored rivals.

Limits

• Value is Unrated — no price and no Artificial Analysis index are published on this board, so there is no cost story here at all.
• It leads no axis. Everything sits in the upper middle; the frontier tiers beat it everywhere and the specialists beat it where they specialise.
• Speed 76 is its weakest axis and rules out the highest-volume patterns.
• Alibaba's release cadence is fast. Qwen3.8 Max listings from September 2026 already sit on this board unscored at $2/$6 and $4/$12 — verify what is current before building on any one version.

Price and access

No published price here — the board lists no rate, no index and no gateway for this model. The unscored Qwen3.8 Max listings on the board do show hosted rates, which is a reasonable indication of the tier's likely cost, but this page measures Qwen 4 Max as of 15 Sep 2026 and no rate was published for it at scoring time.

Alternatives on this board

• GPT-5.6 Terra — 86 at $2/$12 with Context 90, one point up with a known price.
• DeepSeek V4 — 86 at $0.78/$1.57 with Coding 92 and open weights, the cheaper Chinese alternative.
• Grok 5 — 87 with Reasoning 91, two points up if a closed frontier model is acceptable.
• Qwen 4 Coder — 84 with Coding 90, when the workload is code and nothing else.

Sources