
Together raised its Qwen4 Max output price
By Rob Litterst, Co-founder, PricingSaaS
Published
Together raised Qwen4 Max's per-token prices across the board.
Together’s pricing page · Change detected
What changed
According to Together's pricing page, Qwen4 Max list rates rose on Nov 20, 2026, from $1.43 to $1.50 for input, $0.0713 to $0.075 for cached input, and $4.28 to $4.50 for output, all per 1M tokens. Together's pricing page also marks a limited-time discounted rate — $1.05 input, $0.0525 cached and $3.15 output per 1M tokens — as valid only through 11/23/2026. Separately, per Together's pricing page, Together added a new model, Tencent Hy4, to its serverless inference catalog on Nov 20, 2026, at $0.83 input, $0.042 cached and $2.50 output per 1M tokens.
Why it matters
For a product leader evaluating or reselling AI inference, the interesting part isn't the increase itself — it's the timing: Together's Qwen4 Max model was brand new, and per Together's pricing page its list rate had already moved to $1.50 input / $0.075 cached / $4.50 output per 1M tokens, up from $1.43 / $0.0713 / $4.28, as of Nov 20, 2026. Any team that just quoted a customer off Together's rate card, or built Qwen4 Max into its own usage-based pricing, is already behind the vendor's own price. Pulse has tracked 46 real pricing changes from Together since Dec 29, 2025, averaging about one every 7 days, so a wholesale AI-inference rate is less a fixed cost than a moving one to re-check before every quote. Together's pricing page marks its new discounted rate of $1.05 input / $0.0525 cached / $3.15 output per 1M tokens as valid only through 11/23/2026 — watch whether that becomes Together's real price once the window closes, or whether Qwen4 Max gets repriced again before then.
In context
Together's own record makes this look like business as usual: Pulse has tracked 46 real pricing changes from Together since Dec 29, 2025, including a 67% hike on Qwen3.7-Max on Nov 7, 2026 [[together/2026D311]] and a GPU-rate rise to $9.29/hr on Nov 13, 2026 [[together/2026D317]], so repricing a model one day after launch fits the pattern. The sharper read is the two-tier structure: Together is running a visible discount against its own new list price, echoing how, per Pulse's record, it let a Kimi K3 cached-input promo lapse back to $0.30 per 1M tokens on Oct 14, 2026 [[together/2026D287]]. Elsewhere, Pika [[pika]] and Lumaai [[lumaai]] both touched per-generation credit pricing on Nov 19, 2026, and Pulse tracked 416 such credit/usage-billing changes across 276 companies in the last 180 days — a live pattern, but still concentrated in AI-infra and generation tools rather than SaaS broadly. Together's pricing page marks the real signal here: whether its $1.05/$3.15 promo rate, valid only through 11/23/2026, quietly expires the same way.
Common questions
How much more would this cost us per 1M tokens?
Take the delta between Together's old and new Qwen4 Max rates on its pricing page and multiply it by your expected token volume for that model; because the discounted rate is temporary, budget off the full list price shown as of Nov 20, 2026, not the promo one.
Does the record show why Together raised the price the same day the model launched?
No — the pricing page shows the new rate and the discount but gives no stated reason, and there's no on-the-record comment from Together about this change, so treat any explanation as your own read rather than a stated rationale.
Is raising a price one day after launch unusual for an AI inference vendor?
Not really — in the same week Pika and Lumaai both revised their own credit-based pricing, and fast post-launch repricing is common enough across usage-billed AI tools that Pulse's data treats it as a live pattern rather than an outlier for Together specifically.
Sources
- Pulse: Together pricing changes, Nov 20, 2026pricingsaas.com/companies/together/diffs/2026D324
- Together pricing pagetogether.ai/pricing