Baseten adds Reflection Beam and Meta Muse Spark Open, a day after its last model
Baseten's Model APIs catalog gained two new entries: Reflection Beam, priced at $1.30 per 1M input tokens, $0.13 for cached input, and $4.30 for output; and Meta Muse Spark Open, at $1.25 input, $0.15 cached, $4.25 output.
Baseten’s pricing page · Change Detected
WHY IT MATTERS
For teams comparing inference providers on price and model breadth, Baseten's two-new-models-in-two-days pace is worth tracking against its typical ~21-day gap between changes. Watch whether Baseten keeps this faster cadence or returns to its historical pace once the current wave of catalog additions settles.
BY THE NUMBERS
- Second model-catalog addition in Baseten's lineup in 2 days
- 5 real pricing changes since tracking began in July 2026
- Cut DeepSeek V4.1 Flash cached input 77% to $0.007 in September
- 201 credits/usage-billing changes across 153 companies in 180 days
THEIR RATIONALE
Baseten frames these as more pre-optimized models instantly available on its Inference Stack — this is about catalog breadth, not a price change to existing models. It comes just one day after Baseten added Nemotron 3.5 Super VL, making it two model launches in two days against a typical cadence of about 21 days between changes.
THE SIGNAL
Baseten has made 5 real pricing changes since it started being tracked in July 2026, almost all of them tied to credits-and-usage billing — including yesterday's Nemotron 3.5 Super VL addition and a 77% cut to DeepSeek V4.1 Flash's cached input price in September. Catalog speed is becoming a differentiator among inference resellers: 201 pricing changes across 153 companies have touched credits-and-usage billing in the last 180 days, 11.6% of everything tracked, and Baseten is adding rows to its table faster than its own historical pace.
IN CONTEXT
This is Baseten's second model-catalog addition in two days, coming right after Nemotron 3.5 Super VL was added at $0.20 per 1M input tokens on Oct 16 — a sharp acceleration from its typical ~21-day gap between changes. Baseten's other recent moves in this pattern include a 77% cut to DeepSeek V4.1 Flash's cached input price in September and Kimi K3's addition in July. Market-wide, 201 pricing changes across 153 companies have touched credits-and-usage billing in the last 180 days, 11.6% of everything tracked — Baseten's pace this week is faster than most of that cohort.
WHAT PEOPLE ARE SAYING
Muse Spark Open costs $0.84 per Intelligence Index task at current API pricing of $1.25/$4.25 per 1M input/output tokens ($0.15 cached input) from Together AI, Fireworks and Baseten - less than half of GLM-5.3 (max, $2.01), but about three times DeepSeek V4.1 Flash (max, $0.27)
View originalDay one. 7 clouds. One price war. Meta's Muse Spark Open didn't just drop weights yesterday — it launched on 7 inference clouds on day one (Baseten, Fireworks, Groq, Modal, Nebius, OpenRouter, Together AI) plus 4 weight channels (Hugging Face, SageMaker, Model Garden, Foundry). Day-one API price (Together): $1.25/1M input, $4.25/1M output, $0.15/1M cached. The twist: against Kimi K3 ($3/$15), the US open model is suddenly the cheap one — ~$0.84/task vs $2.00 per one index. For years the cost edge ran toward Chinese labs. This pair inverts it. Honest footnote: those prices are a day-one snapshot — all inference goes through third parties, so they'll move as providers compete. The license caps cloud resellers at $20M before Meta wants a deal. The weights were free. The race is over who bills you less to run them.
View originalTogether AI made Muse Spark Open free through Oct 18 (list $1.25 in / $4.25 out per 1M tokens) and added Nemotron 3.5 Super VL at $0.15 / $0.50. @togethercompute: how does this compare with @FireworksAI_HQ and @baseten?
View original