The Bargain Model Tightens Security
Anthropic says Haiku 5.5 costs about 75% less to run than Haiku 4.5 on average, with tighter cyber safeguards. The discount aisle has a stricter bouncer now.
Read the pick: The Bargain Model Tightens SecurityX OPPORTUNITY RADAR / DISPATCH
Developers can call Alibaba’s new flagship through an API. It scored slightly above the median for similarly priced models in one independent test, where its output token use was also above the median.

Alibaba announced the Qwen4-Max launch on November 16, following a limited preview. The company advertises a context window of about one million tokens and offers the model through QwenCloud. Vercel also lists it in its AI Gateway. Developers can use the model through a hosted service, but Alibaba has not released the weights needed to run this version themselves. The company says smaller Qwen4 models will get open weights.[1][3][4]
Alibaba says Qwen4-Max reduced the average cost per task by about 60 percent versus Qwen3.8-Max on Work360, its own benchmark. That is a result for the tasks and settings Alibaba tested. It does not tell a team what the same model will cost on its own work.[1]
Artificial Analysis provides a second measurement. Qwen4-Max scored 43 on its Intelligence Index, against a median of 41 among similarly priced models. The evaluator estimated a weighted average cost of $3.27 per task on that index, using listed prices and the tokens consumed. Across the test, the model generated 127 million output tokens, compared with the evaluation median of 78.6 million.[2]
For teams choosing a model, the number to measure is cost per usable result. A lower price per token can still produce a larger bill if the model needs more tokens to finish the work.
Checked Nov 16, 2026. Numbered references point to the evidence behind factual passages. Commentary is our own.