The state of the art regarding non-interactive bit commitment from OWF is quite simple: it is known to be impossible (at least using black box reductions) to build non-interactive commitments from one-way functions (see e.g. this paper).
Are you specifically interested in constructions from generic one-way function, or are you ok with more structured assumptions, such as one-way permutations (which do imply non-interactive commitments)? If you are also interested in OWP (or even one-to-one OWF), then I think the state-of-the-art generic construction of a NIBC is the direct application of the Goldreich-Levin theorem: to commit to $b$, given any OWP $f:\{0,1\}^n\mapsto \{0,1\}^n$, pick random $(x,r)$ and return $(f(x), r, \langle x, r\rangle \oplus b)$. The communication is $2n+1$ bit per bit committed. If you want to commit to a single bit, I do not know of any better solution. When amortizing, however, there might be better approaches - for example, you can at least commit to $\log(n)$ bits at once using a single $x$ and $\log(n)$ different $r$'s, lowering the communication to $n(1+o(1))$. I don't know if there is anything better generically (getting more hardcore bits from the Goldreich-Levin construction seems hopeless, but perhaps a completely different approach can help).
From collision resistant hash functions, at least, this papers provides a construction of a non-interactive commitment with $O(1)$ amortized communication - but their construction still requires a one-time interactive setup phase, the commitment scheme only becomes non-interactive after that.
