This answer was started with the intention to show the claimed property by using a less involved calculus done by a human hand with a human eye.
However, the complexity cannot be avoided. The only solution that is a fair complete solution is the computer path. It is simple, based on Groebner bases algorithms, and does the job immediately. It also shows why a naive approach cannot lead to simple quick solution. (Except for finding a trick, that i am maybe missing.) So i will also insert a human computational start, and a geometric point of view, also incomplete.
Computer view:
Psychologically it is a benefit to see first the computer solution,
mainly because it makes clear, there is only one way to realize the picture.
Here the human hand is imposing algebraic conditions, letting the computer solve a resulted system of algebraic equations. Its complexity equals the complexity of the stated problem.
How do we translate the picture into equations?
We are placing the solution box with the center in the origin of axes, and let its width and height be $2w$ and $2h$.
Fig. $1$:

The three circles have the radii $a,b,c$, and the upper $b$-radius circle has its $x$-coordinate $x_u$. The code below considers a polynomial ring $R$ in the six free variables $a,b,c;h,w;x_u$, immediately sets the half-height to one, $h=1$, and let $J$ be the ideal generated by the equations which reflect:
- $h=1$,
- upper $a$-circle is tangent to the upper $b$-circle,
- upper $b$-circle is tangent to the upper $c$-circle,
- upper $b$-circle is tangent to the lower $a$-circle,
- upper $c$-circle is tangent to the lower $a$-circle,
- upper $b$-circle passes through the origin
- the line between the two centers of $a$-circles is tangent in the origin to
the line between the two centers of $a$-circles.
There are seven equations, six variables, we check the dimension of the ideal, it is zero, and ask for its "variety" with coordinates in the real algebraic numbers, which is a set of points.
Sage Code: (Please skip if it feels annoying.)
R.<a,b,c,h,w,xu> = QQ[]
O = vector([0, 0,])
Oa = vector([-w + a, h - a])
Ob = vector([ xu, h - b])
Oc = vector([ w - c, h - c])
Oaa, Obb, Occ = -Oa, -Ob, -Oc
J = R.ideal([h - 1
, d2(Oa, Ob) - (a+b)^2
, d2(Ob, Oc) - (b+c)^2
, d2(Oc, Oaa) - (a+c)^2
, d2(Ob, Oaa) - (a+b)^2
, d2(Ob, O) - b^2, (Oa-Oaa)*Ob])
for sol in J.variety(AA):
if sol[w] <= 0: continue
if sol[a] <= 0: continue
if sol[b] <= 0: continue
if sol[c] <= 0: continue
print(f'SOLUTION:')
print(f' a = {sol[a]} with minimal polynomial {sol[a].minpoly()}')
print(f' b = {sol[b]} with minimal polynomial {sol[b].minpoly()}')
print(f' c = {sol[c]} with minimal polynomial {sol[c].minpoly()}')
print(f' w = {sol[w]} with minimal polynomial {sol[w].minpoly()}')
print(f' xu = {sol[xu]} with minimal polynomial {sol[xu].minpoly()}')
And here are the solutions:
SOLUTION:
a = 1 with minimal polynomial x - 1
b = 0.50000000000000000? with minimal polynomial x - 1/2
c = 5.828427124746190? with minimal polynomial x^2 - 6*x + 1
w = 2.414213562373095? with minimal polynomial x^2 - 2*x - 1
xu = 0 with minimal polynomial x
SOLUTION:
a = 0.606577858091834? with minimal polynomial x^6 - 2*x^5 - x^4 + 8*x^3 + 3*x^2 - 6*x + 1
b = 0.5210075620474032? with minimal polynomial x^6 - 2*x^5 - x^4 - x^3 + 2*x^2 - x + 1/4
c = 0.4037108932628932? with minimal polynomial x^6 - 6*x^5 + 23*x^4 - 48*x^3 - 5*x^2 + 6*x + 1
w = 1.525935796142292? with minimal polynomial x^6 + 8*x^4 - 8*x^3 + 16*x^2 - 32*x - 16
xu = 0.2049759110110414? with minimal polynomial x^6 - 2*x^5 + 3*x^4 - 8*x^3 + 7*x^2 - 6*x + 1
The first "solution" found is not a solution to our problem, but a solution of the problem of drawing six circles tangent to each other as specified
and (somehow) to the axes of a box. Here is a plot of it, since it can be easily plotted with exactly specified coordinates.
Fig. $2$:

The missing circle is the reflection in the origin of the big brown $c$-circle.
The second solution was already plotted in the Figure $1$, and the values of $a,b,c;w,x_u$ are algebraic of degree $6$. This complexity cannot be eliminated.
The moral of the story is: We have exactly one configuration realizing the posted OP picture, and when normed to a height $H=2h=2$ it involves
for the radii and the width complicated algebraic numbers (living in algebraic fields of degree six).
Note: Using similar eliminations, the computer shows that for $w\ne 0$ and $h=1$ from the given equations we have "only" $(e^2-1)(e^2-2)=0$. This relation is involved for the human eye, and also confirms the complexity of reaching the wanted $e=1$, after discarding $e=\sqrt 2$.
An algebraic path (incomplete solution): (Done humanly.) We are trying to solve the same system in a human manner:
$$
\left\{
\begin{aligned}
x_u^2 +(h-b)^2 &= b^2 & & \text{i.e.} &d(O_b,O) &= b\\
(x_u + w-a)^2 +(a-b)^2 &= (a+b)^2 & & \text{i.e.} &d(O_a,O_b) &= a+b\\
(-x_u + w-c)^2 +(b-c)^2 &= (b+c)^2 & & \text{i.e.} &d(O_b,O_c) &= b+c\\
(a-c)^2 +(2h-a-c)^2 &= (a+c)^2 & & \text{i.e.} &d(O_c,O'_a) &= a+c\\
x_u(w-a) & = (h-a)(h-b)& & \text{i.e.} & Oa_O&\perp O_bO\\
e^2 &= (a+b)^2 - b^2
\end{aligned}
\right.
$$
Below we are setting $h=1$, and trying to eliminate everything but $e$.
The first variable we eliminate below is $x_u$. We have $x_u=(1-a)(1-b)/(w-a)$, and $x_u^2=2b-1$. So the above system implies:
$$
\left\{
\begin{aligned}
(1-a)^2(1-b)^2 &= (2b-1)(w-a)^2\\
(2b-1) + 2(1-a)(1-b) + (w-a)^2 &= 4ab\\
(2b-1) - 2(1-a)(1-b)\frac{w-c}{w-a} + (w-c)^2 &= 4bc\\
(2-a-c)^2 &= 4ac\\
e^2 &= (a+b)^2 - b^2
\end{aligned}
\right.
$$
We eliminate $w$ by using the expression for $(w-a)^2=(1-a)^2(1-b)^2/(2b-1)$ from the first equation in the second one. Also, we can subtract above the second and third equations from each other to cancel the term in $w^2$.
This gives $\displaystyle 2(1-a)(1-b)\left(2+\frac{a-c}{w-a}\right)-2w(a-c) +(a^2-c^2) =4b(a-c)$. We divide by $(a-c)$ and use the radical formula for $(w-a)$.
$$
\left\{
\begin{aligned}
0 &= a^2b^2 - 2a^2b - 6ab^2 + a^2 + 2ab + b^2 \\
4b &= 2(1-a)(1-b)\left(\frac2{a-c}+\frac{\sqrt{2b-1}}{(1-a)(1-b)}\right)
-2\frac{(1-a)(1-b)}{\sqrt{2b-1}}-a-c\\
(2-a-c)^2 &= 4ac\\
e^2 &= (a+b)^2 - b^2
\end{aligned}
\right.
$$
The equation in $a,c$ only lets us write $c$ as $a+2\pm\sqrt{8a}$.
Because of $c<2$ we have $c=a+2-\sqrt{8a}$. This leads to:
$$
\left\{
\begin{aligned}
0 &= a^2b^2 - 2a^2b - 6ab^2 + a^2 + 2ab + b^2 \\
0 &=
(1-a)(1-b)\cdot\frac1{\sqrt{2a}-1}
+\sqrt{2b-1}
-\frac{(1-a)(1-b)}{\sqrt{2b-1}}
+1
-\sqrt{2a}
-2b\\
e^2 &= (a+b)^2 - b^2
\end{aligned}
\right.
$$
Now the first equation can be written as $0=a^2(1-b)^2-2ab(3b-1)+b^2$,
and we consider it as an equation of second degree in $a$. This gives
$$
a_\pm =\frac b{(1-b)^2}\Big[3b-1\pm2\sqrt{b(2b-1)}\Big]\ .
$$
The right sign is minus, so we have a formula for $a$ in terms of $b$:
$$
a =\frac b{(1-b)^2}\left(\sqrt b -\sqrt{2b-1}\right)^2\ .
$$
In order of making computations without radicals, we are searching for a parametrization of $b$ so that $\sqrt b$ and $\sqrt{2b-1}$ are rational expressions. One such parametrization is
$$
\small
\begin{aligned}
b &= \left(\frac{1+u^2}{1+2u-u^2}\right)^2
&
a &=\frac 14\left(\frac{1+u^2}{1-u}\right)^2
\\
\sqrt b &= \frac{1+u^2}{1+2u-u^2}
&
\sqrt a &= \frac{1+u^2}{2(1-u)}
&
\sqrt{2b-1} &= \frac{1-2u-u^2}{1+2u-u^2}
\\[2mm]
1-a &= \frac {(1-2u-u^2)(3-2u-u^2)}{4(1-u)^2} &
1-b &= \frac {4u(1-u)(1+u)}{{(1+2u-u^2)^2}}\ ,\\[2mm]
e &=\frac 14\cdot\frac{(3-2u+u^2)(1+u^2)^2}{(1+2u-u^2)(1-u)^2}\ ,
\end{aligned}
$$
where $u$ is a suitable real value. Using $u$ in the remained equation (involving radicals, now made to no longer involve them), we obtain:
$$
\begin{aligned}
0 &= \frac{(1-a)(1-b)}{\sqrt{2a}-1}
-\frac{(1-a)(1-b)}{\sqrt{2b-1}}
-\sqrt{2a}
+\sqrt{2b-1}
-(2b-1)\\
&=\frac{(1-a)(1-b)}{\sqrt{2a}-1}
-
\frac {u(1+u)(3-2u-u^2)}{(1-u)(1+2u-u^2)}
-\sqrt{2a}
+
\frac{4u(1-2u-u^2)}{(1+2u-u^2)^2}
\end{aligned}
$$
This can continue, but i have to stop here, the effort does not
compensate for the result.
Geometric approach (incomplete): (But geometric properties are collected.)
We start with the following situation, which is a subconstellation in the given constellation of circles:

Here, the "same" notation $a,b$ is used for the radii of the two pairs of equal (here in the sense of congruent) circles. The "red segment" $AA'$ matches the OP red segment.
Explicitly, we introduce the involved points in the following order, so that everything is well defined. (We use cartesian axes only as orientation, no involved analytic geometry is needed.)
Let $O$ be a point, here taken as the origin of the axes.
Denote by a prime the reflection w.r.t. $O$, so it is a map of points $X\to X'$.
Let $B$ be a point on the $Oy$-axis at distance $b$ from $O$. Let $A$ be a point on the $Ox$-axis at distance $a+b$ from $B$. In particular, its distance $e$ from $O$ satisfies $e^2=(a+b)^2-b^2=a^2$.
(This $e$ has nothing to do with the later point $E$, nor with $\exp(1)$, i am just needing a new letter...)
(The picture was drawn with $OB=b=36$, $a=49$, $AB=a+b=85$, $e=77$. The triangle $\Delta AOB$ is a $36-77-85$ pythagorean triangle. So $a,b$ are perfect squares, just in case...)
Let $D\in AB$ be the tangency point of the circles
$\mathcal A=\operatorname{Circle}(A,e)$ and
$\mathcal B=\operatorname{Circle}(B,b)$.
Construct all reflected objects, the points $A',B',D'$, and the circles $$\mathcal A',$\mathcal B'$.
Reflect $D$ w.r.t. $OB$ to obtain $E$, and build $E'$.
So $D'\in A'B'$ is the tangency point of the circles
$\mathcal A'=\operatorname{Circle}(A',\alpha)$ and
$\mathcal B'=\operatorname{Circle}(B',b)$.
And $E,E'$ are tangency points for the pairs of circles $(\mathcal A',\mathcal B)$ and
$(\mathcal A,\mathcal B')$.
The exterior tangents of $(\mathcal A',\mathcal B)$ intersect in a point $S$, and these tangents are $SJK$, $SLM$, with $J,L\in \mathcal B$, and $K,M\in \mathcal A'$. Build also their reflections.
Arrange in notation of the two tangents that $SJK$ and its reflection $S'J'K'$ delimit a region of the plan (intersection of two half-planes) where all the picture lives in. (And the needed box will be the minimal rectangular box in this strip
between the parallels $SJK\|S'J'K'$.
We will build perpendiculars on them which are tangent to $\mathcal A,\mathcal A'$ on the far side w.r.t. $O$).
(This was a long exact description of what could be omitted, given the picture and a wink with the hands.)
We would want to get the sizes of the box.
As a warm up, let us compute the coordinates of some easy involved points.
$$
\begin{aligned}
B &=(0,b) %& B'&=-B=(0,-b)
\\
A &= (e,0) %& A' &=-A=(-e,0)\ ,
\qquad e^2=(a+b)^2 - b^2 = a(a+2b)\\[2mm]
D &= \frac b{a+b}(e,a) %& D' &= -D = -\frac b{a+b}(e,a)
\\
E &= \frac b{a+b}(-e,a) %& E' &= -E = -\frac b{a+b}(-e,a)= \frac b{a+b}(e,-a)
\\[2mm]
S &= \left(\frac{be}{a-b}\ ,\ \frac{ae}{a-b}\right)
%&S' &= -S =-\left(\frac{be}{a-b}\ ,\ \frac{ae}{a-b}\right)
\end{aligned}
$$
For $D=(x_D,y_D)$ we have used $y_D:b=y_D:OB=a/(a+b)$, and $x_D:b=x_D:DB=AO:AB=e:(a+b)$. For $S=(x_S,y_S)$ we have built similar to $\Delta A'BO$ a similar triangle $\Delta A'SP$, $P$ being the projection of $S$ on $AA'$. Then $y_S:(x_S+e)=SP:A'P=BO:A'O=b:e$, and a second relation can be obtained from the trapezium $A'BJK$. Project $B$ on $A'K$, we obtain a right triangle with sides $A'B=(a+b)$, and catheti $(a-b)=A'K-BJ$, and $2\sqrt {ab}$ (same as $JK$). This triangle is similar to $\Delta SJB$, so $SB:b=SB:JB=(a+b):(a-b)$, leading to a formula for $SB$, which can be used to obtain $x_B$ from the similarity $x_B:SB=e:(a+b)$.
Lemma A:
The (light yellow) circle with diameter $A'B$ is tangent to $SJK$ in the mid point of $JK$, this mid point has same distance $\sqrt {ab}$ to $J,K,E$.
Note: This circle is thus after a reflection w.r.t. $SBA'$ also tangent to $SLM$ in the mid point of $LM$, this mid point has same distance $\sqrt {ab}$ to $L,M,E$.
The two mid points and $E$ are on the perpendicular in $E$ on $SBA'$.

Proof: Consider the trapezium $A'BJK$, and the mid points
Let $\Omega,T$ of $A'B$, $KJ$.
It may help to take a copy of this trapezium, and to paste it in reflection w.r.t. $\Omega$, this reflection is denoted by a star, so we have in the picture a rectangle $JKJ^*K^*$ centered in $\Omega$. We start the proof.
Project $B$ on $A'K$ in $P$.
Consider the triangles $\Delta A'BP$ and $\Delta \Omega TE$ with $\hat \Omega =\hat{A'}$ and the proportionality of adjacent sides,
$$
\frac{\Omega E}{\Omega T}
=
\frac{(a+b)/2}{(a-b)/2}
=
\frac{(a+b)}{(a-b)}
=
\frac{A'B}{A'P}\ .
$$
They are similar, so $\hat E=\hat P=90^\circ$, and $TE=BP/2=KJ/2=\sqrt{ab}=TK=TJ$.
$\square$
Lemma B:
The points $J,L,E'$ from Figure $1$ are collinear.
Proof:
Let $R$ be the intersection of $JL$ with $SBEA'$. Let $O_1$ be the projection of $O$ on this line.
By symmetry it is enough to show $O_1E=O_1R$. So we compute:
$$
\begin{aligned}
BR &=\frac{BJ^2}{BS}=\frac{b^2}{b(a+b)/(a-b)}=b\cdot\frac{a-b}{a+b} \\
O_1B &= \frac{BO^2}{BA'}=\frac{b^2}{a+b}=b\cdot\frac{b}{a+b} \\
O_1R &= O_1B+BR=b\cdot\frac{(a-b)+b}{a+b}=b\cdot\frac{a}{a+b} \\
O_1E &= BE-BO_1 =b-b\cdot\frac{b}{a+b} =b\cdot\frac{a}{a+b}
\end{aligned}
$$
so $O_1R=O_1E$.
$\square$
The height of the box:
Which is the distance $h$ from $O$ to the tangent $SJK$? Its double $H:=2h$ is the height
of our box.
Recall that $O$ is one intersection point of the green circle $\odot(B)$
with the yellow circle $\odot(\Omega)$. We denote by $\gamma,\eta$ the angles
$\gamma=\widehat{A'SK}=\widehat{PBA'}$, and $\eta=\widehat{OA'B}$.
Then $\widehat{OBK^*}=\eta-\gamma$.
The distance $h$ from $O$ to the tangent $SJK$ is then $JB=b$ plus the projection of $OB=b$ on the side $BK^*$, which is $b\cos(\eta-\gamma)$.
If we really want, an algebraic formula for $h$ can be written,
$$
\begin{aligned}
h &=b+ b\cos(\eta-\gamma)=b+b\cos\eta\cos \gamma+b\sin\eta\sin\gamma
\\
&=
b
+b\cdot \frac e{a+b}\cdot\frac{2\sqrt{ab}}{a+b}
+b\cdot \frac b{a+b}\cdot\frac{a-b}{a+b}
\ ,
\end{aligned}
$$
which can be simplified, but a geometric view may want to work with $\eta,\gamma$.
The width of the box:
Which is the width $W=2w$ of the box? For this we draw a parallel to $SJK$
through $A'$. Let $Q$ be its far most intersection with the circle $\mathcal A'$, seen from $O$.

Then the half-width $w$ is the projection of $OQ$ on the direction $SJK$,
$$
\begin{aligned}
w &= QA'+PB-OB\sin(\eta-\gamma)
= a + 2\sqrt{ab} - b\sin(\eta-\gamma)\ ,\text{ so}\\
w-a &= 2\sqrt{ab}\cdot\frac{e^2}{(a+b)^2} + e\cdot\frac{b(a-b)}{(a+b)^2}\ .
\end{aligned}
$$
and again, an algebraic formula in $a,b$ can be given.
The corner tangencies:
We come closer to the fully filled box, the question is when a further circle of radius $c$ in OP notation may fit in the corner of the box, so that it is tangent to the two box sides at the corner, and to both a green circle and a red circle.
We are in the following situation:

The equations extracted from the picture are:
$$
\left\{
\begin{aligned}
(a-c)^2 +(t-c)^2 &= (a+c)^2\\
(b-c)^2 +(s-c)^2 &= (b+c)^2\\
(a-s)^2 + (b-t)^2 &= (a+b)^2
\end{aligned}
\right.
$$
and in our case the lengths of $s,t$ are:
$$
\left\{
\begin{aligned}
s &=w-b\sin(\eta-\gamma)\\
t &=h-a\ .
\end{aligned}
\right.
$$
One can simplify to some extent, but again, algebraic computations of a higher degree are involved.