Just a simple question...

I know there is testing going on, but what is the curernt recommendation?

From what little I've read, I (note that I am not an expert) would use 3.96 for CBR and ABR and 3.90.3 for VBR. I have no idea if this is the consensus though, just my opinion.

Tests of 3.96 are still going on and haven't reached a conclusion. As far as I know, 3.90.3 is still 'recommended' but besides a number of minor issues, there doesn't seem to be anything seriously wrong with 3.96.

edit:got my lame versions wrong.

Don't you mean 3.90.3?

Tests have not yet concluded on LAME 3.96 to determine whether it should be become the recommended version or not. Though it depends on who you ask when it comes to which version to use. I decided to upgrade to LAME 3.96 because the quality is good to my ears and it provides lower bit rates and faster encoding. Keep in mind that LAME 3.90.3 (the current recommended version) is the most tested and tuned version of LAME and is why it is recommended here. This is why you see a lot of people using it and telling others to use it. However, I personally like progress and the latest version of LAME suits me just fine so I'm using it. Use what sounds best to you on your music and be happy.

3.90.3 is still the Hydrogenaudio.org approved and recommended version. Once that changes it will be announced (and celebrated ) and the sticky thread in the MP3 - General forum will be edited accordingly.

Testing is going on and if you want to help improving lame be sure to participate in this test.

The only 2 settings with a mentionable number of test results so far are --(alt-)preset standard and ~128kbps ABR/VBR. Details:

--(alt-)preset standard:
In the tests with 3.96beta1 here 3.90.3 won on 13/17 samples. This would be a "p-value" of 2.5%. ~ 50% of the samples tested were known lame 3.90.3 problem samples so one could assume a slight bias against 3.90.3. OTH the samples didn't really cover a broad/representative variety of music. Anyway, 3.96 would need to be better on at least 50% of samples tested - and it's a long way till there, no matter what type of music samples tested in future might be.
The only change between 3.96 beta1 and final for --preset standard was the minimum bitrate set to 128kbps. The 4 samples that were tested after this change were still worse then 3.90.3.

Conclusion: Surely it's too early to tell, but it's likely that 3.96 isn't better then 3.90.3 with --(alt-)preset standard setting, probably it's worse - so for this setting 3.90.3 has good chances to stay the recommended version.

~128kbps ABR/VBR:
Here it's the other way round. It looks like 3.96 with --preset 128 but also with -V 5 are better then 3.90.3 --alt-preset 128. If --preset 128 or -V 5 will win (and if there'll be a clear winner at all) is impossible to estimate right now.

I actually just did some ABX tests and it does seem that 3.90.3 is still better with --alt-preset standard. Though for the bitrate reduction LAME 3.96 doesn't degrade the quality too much. Since I can ABX I'm just sticking with 3.90.3 for the time being but I think the LAME devs are making progress in improving a nearly perfect encoder in 3.90.3. It's not an easy task to try to improve something that is as highly tuned as 3.90.3. With the help of the HA community I think eventually we will be able to replace the 3.90.3 build as the recommended version.

By the way, what should I use to participate in the LAME 3.96 vs 3.90.3 listening tests? Where can I download samples?

http://ff123.net/samples.html

How would you feel about making a change?
We fear change.