Disclosure: I'm a Blockstream co-founder

Major feature contributions to Bitcoin Core:

• Greg Maxwell, Andrew Poelstra and myself (as well as several other contributors) worked on libsecp256k1, a well-tested library for efficient elliptic curve cryptography using the curve secp256k1. Bitcoin Core switched signing to it in 0.10 and verification in 0.12 (with a 5x speedup for signature validation as a result). This work also led to the discovery of CVE-2014-3570 (OpenSSL bignum squaring bug). Work on the library started as a hobby project of mine in 2013, but many optimizations, tests, correctness proofs, and automated verification were added after we could work on it full time.

• I worked on headers-first synchronization (Bitcoin Core 0.10), a new mechanism for block chain downloading which downloads from multiple peers at once, and solved many of the problems with the earlier mechanism. The idea was first described by Greg Maxwell in 2012, but the implementation only got finished after starting at Blockstream.

• Jorge Timon and Matt Corallo (as well as Cory Fields) worked on the creation of libbitcoinconsensus, a shared library built from the Bitcoin Core codebase that exposes part of the validation logic, and shipped with 0.10.

Besides that, we provided a large number of changes implementing numerous features, cleanups, optimizations, refactors, improvements, and general maintenance. When looking at the history of Bitcoin Core v0.12.0rc1 after august 1st 2014, 500 out of 2185 commits (23%; over 30% when including libsecp256k1) came from Blockstream employees and contractors. Most of these people were active in the project long before the creation of the company, as these same people were responsible for 866 out of the 4195 commits before august 1st 2014 (20%). These numbers exclude merges, to avoid favouring people with commit access to the repository.

We were also involved in several improvements not directly related to Bitcoin Core code changes:

• BIP66 fixed a potential forking risk from the network that was caused by an inconsistency across platforms between OpenSSL versions and opened the door to non-OpenSSL based validation.
• BIP68, BIP112, and BIP113 were proposed and first implemented by Mark Friedenbach to introduce relative locktimes and enforcement thereof, which are necessary for more efficient payment channel systems like Lightning.
• Several generic blockchain technology improvements were discovered by us and implemented in our first technology demo sidechain Elements Alpha, including segregated witness, confidential transactions, Schnorr signatures and key tree signatures. Segregated witness is currently being worked on as a soft-fork for Bitcoin (BIP141 through BIP144), and includes script versioning which makes introducing things like Schnorr signatures in future softforks much easier.
• Other proposed BIPs: BIP99 (consensus change best practices) by Jorge Timon, BIP103 (block size increases following technological growth) by me, BIP111 (NODE_BLOOM flag) by Matt Corallo, BIP9 (Version bits with timeout and delay) by me, Greg Maxwell and Rusty Russell (as well as Peter Todd).
• Matt Corallo implemented and operates the Bitcoin relay network, providing low latency block propagation to various parties.

This list is by no means exhaustive, and I'm probably focusing mostly on contributions I participated in myself.

I know of their work on:

• Sidechains and the soon to be released Liquid Sidechain and,
• the Lightning Network

They lead these efforts and are close to implement a production sidechain. These contributions are quite impressive.

For the above projects some changes are required to bitcoin-core as well (e.g. OP_CHECKLOCKTIMEVERIFY, OP_CHECKSEQUENCEVERIFY, ...) and they contribute towards that end as well.