Together added Tev1 4B Experimental to its pricing table
By Rob Litterst, Co-founder, PricingSaaS
Published
Together added Tev1 4B Experimental at $0.04 per million input tokens with free output.
Together’s pricing page · Change detected
What changed
The pricing page Pulse saved on Oct 2, 2026 shows Together added a new Tev1 4B Experimental model to its serverless inference table, priced at $0.04 per 1M input tokens with output tokens free. The addition joins a roster of Qwen and DeepSeek models that Together has repeatedly repriced in recent months.
Why it matters
For a product leader evaluating inference vendors, the signal here isn't the price tag so much as the shape of it: a near-free entry point for a new, smaller model is a cheap way for Together to get developers trying it before it settles on a standard price. If Tev1 4B Experimental proves useful for lightweight tasks, expect its free-output terms to be temporary — Together has a track record of revising prices once a model moves past the experimental label.
Watch whether Together publishes a standard, non-experimental price for Tev1 4B in the coming weeks, and model your own evaluation budget assuming free output tokens won't last.
In context
Pulse has tracked 38 real pricing changes from Together since Dec 29, 2025, and this addition follows a familiar rhythm: its last move came just 7 days earlier, when it reveals NVIDIA HGX B300 pricing on Sep 25, 2026. Most of those changes have hit Together's flagship Qwen models with repeated price increases, so adding a new experimental model with free output tokens reads as Together hedging toward lower-cost options while it raises prices elsewhere. The move also fits a broader pattern of AI and SaaS vendors leaning on credits and usage-based pricing to lower the cost of trying something new; on the same day, Synthesia moves to credits, cuts free plan allowance. For a product leader benchmarking inference vendors, the free-output tier is the detail worth watching — it's a cheap way for Together to get a new model into developers' hands before deciding how to price it long-term.
Common questions
How does $0.04 input pricing compare to Together's other models?
Together's flagship Qwen3.7-Max model currently prices input at $1.50 per 1M tokens after a Sep 23, 2026 price cut, per Pulse's record. Tev1 4B Experimental is priced far below that, which fits Together's pattern of pricing new, smaller models well under its flagship line.
Is Together's pricing changing often lately?
Pulse recorded 20 pricing changes from Together in the last 90 days, including several price increases on Qwen3.7-Max, so new model additions like this one are arriving alongside a steady stream of repricing elsewhere, not in isolation.
Does this signal where Together's pricing is headed next?
Not necessarily — Together has mixed cheap, experimental additions with price increases on its existing models, including a 25% hike to Qwen3.7-Max pricing on Sep 16, 2026, according to Pulse's record, so there's no single direction to read from one new model.
What people are saying
“A classifier that cost $17 to train and serves at $0.042/M input. That is the real story, not the model itself. Open weights are now cheap enough to iterate like software. Benchmarking this against a full-size classifier this week.”
View original“wow, $17 to train and $0/M output makes judge cost stop being a design constraint.... nice. The number I'd want before swapping: tev1-4B agreement with human labels vs a frontier judge on the same rubric. Cheap judging only helps if the agreement gap is small.”
View original“tev1-4B, together's jev-like classifier on qwen3.5 4B: ~$17 to train, ~25 min, 37,840 examples. $0.042/1M input, output free. the catch is buried in the tutorial: call the public endpoint directly and you have to set temperature=0, max_tokens=8 and thinking off yourself. it doesn't inject them”
View originalSources
- Pulse: Together pricing changes, Oct 2, 2026pricingsaas.com/companies/together/diffs/2026D275
- Together pricing pagetogether.ai/pricing