diff --git a/Data/ByteString/Builder/RealFloat.hs b/Data/ByteString/Builder/RealFloat.hs
index 1fef16a0b..9035ae157 100644
--- a/Data/ByteString/Builder/RealFloat.hs
+++ b/Data/ByteString/Builder/RealFloat.hs
@@ -68,6 +68,7 @@ module Data.ByteString.Builder.RealFloat
   , standardDefaultPrecision
   , scientific
   , generic
+  , shortest
   ) where
 
 import Data.ByteString.Builder.Internal (Builder)
@@ -128,11 +129,18 @@ scientific = MkFloatFormat FScientific Nothing
 generic :: FloatFormat
 generic = MkFloatFormat FGeneric Nothing
 
+-- | Standard or scientific notation depending on which uses the least number of charabers.
+--
+-- @since ????
+shortest :: FloatFormat
+shortest = MkFloatFormat FShortest Nothing
+
 -- | ByteString float-to-string format
 data FormatMode
   = FScientific     -- ^ scientific notation
   | FStandard       -- ^ standard notation with `Maybe Int` digits after the decimal
   | FGeneric        -- ^ dispatches to scientific or standard notation based on the exponent
+  | FShortest       -- ^ the actual shortest length string of either standard notation or scientific notation
   deriving Show
 
 -- TODO: support precision argument for FGeneric and FScientific
@@ -163,8 +171,10 @@ data FormatMode
 formatFloat :: FloatFormat -> Float -> Builder
 formatFloat (MkFloatFormat fmt prec) = \f ->
   let (RF.FloatingDecimal m e) = RF.f2Intermediate f
-      e' = R.int32ToInt e + R.decimalLength9 m in
-  case fmt of
+      e' = e'' + olength
+      e'' = R.int32ToInt e
+      olength = R.decimalLength9 m
+  in case fmt of
     FGeneric ->
       case specialStr f of
         Just b -> b
@@ -177,6 +187,14 @@ formatFloat (MkFloatFormat fmt prec) = \f ->
       case specialStr f of
         Just b -> b
         Nothing -> sign f `mappend` showStandard (R.word32ToWord64 m) e' prec
+    FShortest ->
+      case specialStr f of
+        Just b -> b
+        Nothing ->
+          if    e'' >= 0 && (olength + 2   >= e''  || olength == 1 && e'' <= 2)
+             || e'' <  0 && (olength + e'' >= (-3) || olength == 1 && e'' >= (-2))
+           then sign f `mappend` showStandard (R.word32ToWord64 m) e' prec
+           else BP.primBounded (R.toCharsScientific (f < 0) m e) ()
 
 -- TODO: support precision argument for FGeneric and FScientific
 -- | Returns a rendered Double. Returns the \'shortest\' representation in
@@ -206,8 +224,10 @@ formatFloat (MkFloatFormat fmt prec) = \f ->
 formatDouble :: FloatFormat -> Double -> Builder
 formatDouble (MkFloatFormat fmt prec) = \f ->
   let (RD.FloatingDecimal m e) = RD.d2Intermediate f
-      e' = R.int32ToInt e + R.decimalLength17 m in
-  case fmt of
+      e' = e'' + olength
+      e'' = R.int32ToInt e
+      olength = R.decimalLength17 m
+  in case fmt of
     FGeneric ->
       case specialStr f of
         Just b -> b
@@ -220,6 +240,14 @@ formatDouble (MkFloatFormat fmt prec) = \f ->
       case specialStr f of
         Just b -> b
         Nothing -> sign f `mappend` showStandard m e' prec
+    FShortest ->
+      case specialStr f of
+        Just b -> b
+        Nothing ->
+          if    e'' >= 0 && (olength + 2   >= e''  || olength == 1 && e'' <= 2)
+             || e'' <  0 && (olength + e'' >= (-3) || olength == 1 && e'' >= (-2))
+           then sign f `mappend` showStandard m e' prec
+           else BP.primBounded (R.toCharsScientific (f < 0) m e) ()
 
 -- | Char7 encode a 'Char'.
 {-# INLINE char7 #-}