Skip to content
PricingSaaS
Sign in
Back
Pricing briefing
MMindsDB
MindsDB
PRICE INCREASE

MindsDB raised DeepSeek V4.1 Flash pricing and added Claude Sonnet

By , Co-founder, PricingSaaS

Published

MindsDB raised DeepSeek V4.1 Flash output pricing 82% and added Claude Sonnet 5.5.

See before and after

MindsDB’s pricing page · Change detected

What changed

MindsDB raised per-token rates for DeepSeek V4.1 Flash (served via Fireworks AI) on its MindsHub rate card as of the change Pulse detected on Oct 2, 2026: input rose from $0.22 to $0.30 per 1M tokens, output climbed 82% from $0.66 to $1.20, and cache-write pricing increased from $0.22 to $0.30, while cached-input pricing was cut from $0.01 to $0.006 per 1M tokens.

The same rate-card update, per MindsDB's rate card as of Oct 2, 2026, added Claude Sonnet 5.5 (Anthropic) at $2.00 input, $10.00 output, $0.20 cached input and $2.50 cache write per 1M tokens, and removed GLM 5.2 (Fireworks AI), which had been priced at $1.40 input, $4.40 output, $0.14 cached input and $1.40 cache write per 1M tokens; GLM 5.3 remains on the card. According to Pulse's record, the change landed four days after MindsDB's Sep 28, 2026 update that added Opus 5.5 and GPT-6 models and cut both DeepSeek V4 tiers.

Why it matters

For a product leader managing AI infra spend, this is a reminder that token-resale rate cards move independently of the underlying providers' own list prices: MindsDB sets its own markup on Fireworks-hosted and Anthropic-hosted models, and that markup can jump sharply in a single update. An 82% jump in DeepSeek V4.1 Flash's output price, per MindsDB's rate card as of Oct 2, 2026, lands in the same update that adds Claude Sonnet 5.5 at Anthropic-listed rates — MindsDB is repricing its catalog model-by-model rather than holding one flat markup across providers.

Teams routing production traffic through MindsDB's rate card should treat each model line as independently priced and re-check cost assumptions after every update rather than assume a stable discount versus the underlying provider's price. Watch whether MindsDB keeps pairing price increases with new-model additions: this is MindsDB's second price increase since its pricing page was first tracked on Mar 9, 2026, according to Pulse's record.

In context

MindsDB has repriced or restructured its MindsHub rate card repeatedly since it was first tracked on Mar 9, 2026, with 23 real pricing changes recorded so far, including hikes Blaze rates up to 183%, loosens Air free-tier cap on Sep 15, 2026 and adds Opus 5.5 and GPT-6 models, cuts both DeepSeek V4 tiers on Sep 28, 2026, just four days before this update. Pairing a price increase with a new-model addition is becoming MindsDB's pattern rather than the exception. Elsewhere in AI tooling, usage-based repricing is common this cycle: Synthesia moves to credits, cuts free plan allowance by 58% to 500 a month, per Pulse's tracking on Oct 2, 2026 — the same day MindsDB raised its DeepSeek output pricing. For a buyer, the sharper read is that MindsDB's rate card is now a moving target model-by-model, not a stable markup grid — worth re-checking on every update rather than assuming last month's numbers still hold.

Common questions

How much more would heavy DeepSeek V4.1 Flash output usage cost now?

Multiply your monthly DeepSeek V4.1 Flash output-token volume by the new rate on MindsDB's rate card to estimate the cost increase, then compare it against the rate MindsDB lists for the newly added Claude Sonnet line before deciding where to route that traffic.

What should teams relying on GLM 5.2 through MindsDB do now?

Migrate off GLM 5.2 before any cached requests expire — MindsDB dropped it entirely from the rate card in this update, and GLM 5.3 is the closest remaining option still on the card.

How would Claude Sonnet 5.5 cost scale for a team evaluating it on MindsDB?

Check your own ratio of input to output tokens before estimating cost, since Claude Sonnet 5.5 and DeepSeek V4.1 Flash scale very differently per token type on MindsDB's card — the same traffic volume can land at a very different total depending on which model carries it.

Sources