G4 is the latest version of this family of composite methods. In it is included the extrapolation of the limit of the basis set. It is the best and also most expensive, so a lot depends on their molecules.

We also advise you to look for methods of family Wn. They are more computationally expensive, but more complete.

But for TM some theories don't have basis set.

My suggestion to you is to place your question to one of the G4 method developers, namely Prof. Krishnan Raghavachari ([email protected]). The paper "Gaussian-G4 theory" provides some data on Table V as a first information of CPU times. G4 theory was applied to large organic molecules using Gaussian-03 which is another reference where you can e-mail to the authors asking about CPU time: see Struc. Chem.(2011) 22:1303-1314 - "Assessment of Gaussian-4 theory for the computation of enthalpies of formation of large organic molecules" (Olga V. Dorofeeva et al.)

i happened to worked with prof raghavachari recently on a few problematic 1st-row transition metal species (CrOx) using composite methods [JCTC (2012), 8, 3159]. depending on the multireference character of your particular complex, which can often be quite substantial for 1st-row TMs, single-reference composite methods may not work at all, and you might even be better off sticking with DFT. see also JPCA (2009), 113, 5170 [G4(rel)] and JPCA (2012), 116, 870 [ccCA-TM], which provide two composite methods that are supposedly ok for 1st-row TMs.

The current most resorted method is CCD or CCD(T) because of its acuracy, it compares with CAS (complete active space) PT2 = MP2.

See "Excitation energies for transition metal atoms - A comparison between coupled cluster methods and second-order perturbation theory"

Author/s Juraj Raab, Björn Roos, (URL) http://dx.doi.org/10.1016/S006.

Advances in Quantum Chemistry , (2005), v. 48, Pp. 421 - 433

Besides you will need a very good basis set ccQZV.

Should you need large basis set look at EMSL (bse.pnl.gov/‎)

In case you are looking for a heavy atom, here is a reference, where they used CASCF as I said:

"Chemistry. 2012 Jul 23;18(30):9189-93. doi: 10.1002/chem.201200731. Epub 2012 Jun 20.

"A theoretical revisit to the lowest ionization potentials in metal-metal bonds"

Chang IY, Yu JS.

Abstract

The ties that bind: The electrostatic potential of quadruply bonded W(2)(hpp)(4) and its cation support the idea that the very low ionization energies are due to excitations from the metal-metal bonding core. Theoretical MRCI computations based on CASSCF orbitals show that the relativistic effect influences the lowest four ionization energies, leading to distinct IE sequences: IE(δ) < IE(π) < IE(σ) in W(2)(hpp)(4), and IE(δ) < IE(σ) < IE(π) in Mo(2)(hpp)(4).

What software are you using?

I am using GAMESS-US and NWChem.... I can also use ORCA... but as I checked these programs, it seems that the composite methods I mentioned above are not available. The NWChem already had the composite method ccCA.... I am still looking for other free software which can do G3 or G4.... I am thinking of doing the individual calculations involve in the composite method but this could be a tedious work to do...

sure you can using Gamess, Input Description

$CONTRL

runtyp = G3MP2

It does a geometry optimization and the CCSD with its own basis set. Example 43. In terms of time and memory, this method can´t take a whole protein. It has been tested in small ligands. Remember G3 uses large basis sets, aug-ccTZV.

I believe that running Wintendo-7 you are able to work with Firefly-gamess an improved gamess with Granovski additions (I am taking of MP4...)

here is his home page:

http://classic.chem.msu.su/gran/gamess/downloads.html

It is a little different ....

The GAMESS RUNTYP=COMP can run most Composite methods. The code currently has only a few hard-coded methods like ccCA-S4, and ccCA-CCL. However, the code allows the user to request most streams of (standard; unlike FMO) calculations by the use of a fairly detailed set of variables explained in the manual, test cases, and comp.src.

I recommend that you read the article "A study of the rotational barriers for some organic compounds using the G3 and G3CEP theories"-J Mol Model (2014) 20:2199.Douglas Henrique Pereira et al.

Maybe this approach meets your task  requiring less computational cost. With the method G3CEP they got 1.29 kcal.mol-1 deviation against 1.16 kcal.mol-1 deviation with the G3 method. The study contemplates enthalpy of formation, atomization

energy, electron affinity, ionization potential, and proton affinity. Computational costs were reduced from 7% to 70% depending on the molecule's size. 

Try the method out repeting the procedure with a couple of molecules so you get  some feeling about  it in order to proceed with the method or not. Good luck.

.