Cohere cuts Model Vault Small to $3/hr as Embed 5 replaces Embed 4

WHY IT MATTERS

For buyers on Model Vault's Small tier, the move is a straightforward win — same tier name, lower rate, newer model. For the category, it's a reminder that in AI infrastructure pricing, model-generation swaps are becoming the quiet vehicle for price cuts, letting vendors lower headline rates without touching the rest of the table. Watch whether Cohere extends the same Embed 5 economics to Medium and Large tiers, or whether $3/hr becomes the new Small-tier floor across the Model Vault lineup.

BY THE NUMBERS

• Model Vault Small: $4.00 to $3.00/hr, a 25% cut
• 3rd price cut in Cohere's tracked history (since Jan 2024)
• 31 days since Cohere's previous change (Parse 5 tiers)
• Only 3.4% of tracked changes in the last 180 days were price cuts

THEIR RATIONALE

This is Cohere's usual cadence, not a one-off: the company has made 3 real pricing changes in the last 90 days and 9 in the last year, averaging a move every 64 days. The last one was just 31 days ago, adding a Parse 5 doc-parsing model with Medium and XL tiers. Swapping in a faster, cheaper embedding model while holding Medium steady reads as Cohere using a model refresh to sharpen its entry-level price rather than cutting broadly.

THE SIGNAL

Price cuts are a minority move in the index: only 39 of 1,139 tracked changes in the last 180 days (3.4%) involved a price decrease, across 33 companies, and none did it more than twice. Cohere's own history with cuts is thin — this is just its third, after two 2024 token-price reductions on Command R. What makes this one different is the mechanism: Cohere isn't just lowering a price, it's using a model generation swap (Embed 4 to Embed 5) to deliver the cut, the same pattern as inference-cost-driven price cuts elsewhere in AI infrastructure this year, like Baseten's DeepSeek cache-price cut after a model refresh.

IN CONTEXT

Cohere has made this kind of move before but rarely — this is only its third tracked price cut, after two 2024 reductions to Command R's token pricing [[cohere/2024Q3]] [[cohere/2024Q1]]. What's new is the vehicle: instead of repricing an existing model, Cohere retired Embed 4 and used the Embed 5 launch to land a lower Small-tier rate, 31 days after it added Parse 5 tiers to the same Model Vault lineup [[cohere/2026W36]]. That continues a steady cadence — nine changes in the last year — rather than a reaction to market pressure. Price cuts remain rare across the index: just 39 of 1,139 tracked changes in the last 180 days, led by names like Baseten, which cut DeepSeek cache pricing after its own model refresh [[baseten/2026W40]], and Bettermode, which cut its Starter plan 63% [[bettermode/2026W40]].