Together undoes its Qwen3.8 Flash discount while cutting Kimi K3 10%
Together cut Kimi K3 serverless inference pricing 10%, to $2.70 input / $13.50 output per 1M tokens (from $3.00 / $15.00).
Together’s pricing page · Change Detected
WHY IT MATTERS
Buyers benchmarking inference costs on list price should treat Together's cuts as provisional rather than locked in. Watch whether Qwen3.8 Flash holds at $0.15/$0.47 or gets cut again within weeks, which its roughly 7-day change cadence makes plausible.
BY THE NUMBERS
- Qwen3.8 Flash is back to $0.15/$0.47 per 1M tokens, exactly where it stood before Together's Sept. 23 cut 13 days ago
- Together has logged 40 real pricing changes since Dec. 29, 2025 — about one every 7 days
- This is Together's second change in two days; Oct. 5 added the Tev1 4B Experimental model
- Price increases account for 14 of Together's last 40 tracked changes
THEIR RATIONALE
Together doesn't publish a rationale for individual serverless price edits; it treats its model catalog as a running price list it adjusts frequently. This is the company's second change in as many days (Oct. 5 added the Tev1 4B Experimental model) and its 40th tracked pricing change since Dec. 29, roughly one every 7 days, so a token-price tweak like this one fits its routine cadence rather than reading as a one-off announcement.
THE SIGNAL
Qwen3.8 Flash is now priced exactly where it was before the Sept. 23 cut, which suggests that cut was a short-lived competitive move rather than a durable reprice. We ran an experiment-marker probe against Together's live pricing page and found no A/B-test signals, so this reads as a genuine reversal rather than one arm of a live test. It also fits a broader pattern on Together's page: price increases make up 14 of its last 40 tracked changes, including four separate hikes to Qwen3.7-Max since April, even as it discounts newer models like Kimi K3 — selective cuts on the way in, selective hikes once a model is established.
IN CONTEXT
Together's Qwen3.8 Flash price is right back where it started: the company cut it 40% to $0.09/$0.28 per 1M tokens on Sept. 23, and today's change restores the original $0.15/$0.47. That round trip comes amid a broader run of edits to the same handful of models — Together raised Qwen3.7-Max four times between April and September, most recently on Sept. 21, while discounting newer entrants like Kimi K3. Together has now made 40 real pricing changes since it started being tracked in December, averaging one every 7 days, making it one of the most active repricers in the credits-and-usage-billing lens alongside Lumaai and Katteb. The pattern reads less like a stable price list and more like continuous, demand-driven tuning across its serverless catalog.