Devstral Small 2505 vs Llama 4 Scout
Prices checked
Devstral Small 2505 takes 1 of 5 rounds against Llama 4 Scout. It is also the cheaper of the two to run, at $0.06 per million tokens against $0.18.
Mistral AIWins 1 roundDevstral Small 2505
128k context · tool calling · open weights
$0.06/M tokensCheapest price, at NanoGPT · 4¢ per 1,000 pages
MetaLlama 4 Scout
128k context · tool calling · reads images · open weights
$0.18/M tokensCheapest price, at NanoGPT · 7.9¢ per 1,000 pages
Maker's price in (lower wins)
$0.10/M
$0.10/M
Maker's price out (lower wins)
$0.30/M
$0.30/M
Cheapest price (lower wins)
$0.06/M at NanoGPT
$0.18/M at NanoGPT
Maker's price, blended (lower wins)
$0.15/M
$0.15/M
Cost per 1,000 pages (lower wins)
4¢
7.9¢
Context
128k tokens
128k tokens
Open weights
yes
yes
Tool calling
yes
yes
Structured output
not published
not published
Reads images
no
yes
Reasoning
no
no
Released
7 May 2025
5 Apr 2025
Providers selling it
10
19
Price we measure against
the maker's own
OpenRouter
Trains on your prompts (no is better)
not published
not published
Providers selling both · 3
NanoGPT$0.06 / $0.18
TrustedRouter$0.16 / $0.16
Eden AI$0.15 / $0.41
Devstral Small 2505, then Llama 4 Scout, per million tokens.
Go further
Add a third or fourth model, or change the weighting, in the tool.
What moved this week
- $0.10 / $0.30 $0.18 / $0.59
OpenRouter · 47% cheaper · $0.10 / $0.30 $0.18 / $0.59
Kilo Gateway · 20% cheaper · 8¢ / $0.24 $0.10 / $0.30
Frequently asked
- Which is cheaper, Devstral Small 2505 or Llama 4 Scout?
- Devstral Small 2505, at $0.06 per million tokens against $0.18 for Llama 4 Scout, blended 3 in : 1 out, read on 21 Oct 2026.
- Where is each one cheapest?
- Devstral Small 2505: $0.06 at NanoGPT. Llama 4 Scout: $0.18 at NanoGPT.
- Can I buy both from one provider?
- Yes. 3 providers sell both: NanoGPT, TrustedRouter, Eden AI. The cheapest of them for the two together is NanoGPT.
- Which holds more context?
- Both hold 128k tokens in one call.
Prices exclude tax, read 21 Oct 2026 from each provider's own catalog. They are per M tokens, blended 3 in : 1 out, and a page is 600 tokens in and 60 out. Bars are scaled to the larger value in each row.
Related comparisons