Starknet pushed version 0.14.4 to mainnet on October 5, 2026, raising the SNIP-36 proof ceiling to 1.1 billion Layer 2 gas per transaction — roughly the scale of a full Starknet block.

The Ethereum Layer 2 network, built on a STARK-based cryptographic proof system, activated the upgrade after a three-week testnet window that opened September 15, 2026. The defining change is a capacity expansion, not a structural pivot, and it sits inside a multi-year roadmap focused on performance, fee refinement, and decentralization.

What does SNIP-36 actually do?

SNIP-36, a Starknet improvement proposal, governs how developers prove off-chain work. Under the new release, compute-heavy logic runs off-chain and gets submitted as a single proven transaction. Developers pay proof fees rather than full execution fees for every step.

That ceiling has risen substantially. Eligible single-transaction virtual blocks can now reach up to 1.1B Layer 2 gas, a multiple of the prior limit and approximately equivalent to one full Starknet block. The change shifts the cost calculus for any application built around heavy computation that previously had to split work across several transactions.

How do the gas tweaks affect users?

Starknet also adjusted gas weights — the coefficients that price different operations against the actual economic cost borne by the sequencer. Based on recent traffic data, the team estimates these adjustments will shift costs by less than 1% for the majority of transactions.

The previous release, v0.14.3, introduced dynamic L2 gas fees. Version 0.14.4 layers fine-tuning on top of that system rather than redesigning it. For most users, the pricing impact will be marginal.

What breaks if operators don't update?

Two pieces of housekeeping carry operational risk. First, Starknet deprecated legacy feeder gateway endpoints, including call and get_storage_at. Applications still leaning on those endpoints will need to migrate.

Second, and more urgent, node operators face a hard requirement. Full nodes must run Pathfinder v0.24.0 or Juno v0.16.6 to keep operating on the upgraded network. Proofs generated under v0.14.3 are rejected by the new verifier, so outdated setups will not limp along.

The release also moved the block committer into Starknet's Apollo processes, an internal sequencer change with limited external surface area but clear implications for anyone running or auditing sequencer infrastructure.

What changes for STRK and the wider roadmap?

For developers, the SNIP-36 expansion is the headline. For node operators, the client-version requirement is. For holders of STRK, Starknet's native token, the upgrade cycle remains the most concrete signal of network progress, since it ties directly to how fees are structured.

The broader roadmap is unchanged in direction. Starknet continues to work through three stated goals: improving performance, refining fee structures, and moving toward decentralization. The 0.14.x line is incremental on all three axes.

What comes next?

Node operators have a hard clock. Anything not running Pathfinder v0.24.0 or Juno v0.16.6 will stop processing blocks once the new verifier rejects their proofs. Developers on the legacy feeder endpoints face a near-term migration deadline before dependent tooling breaks. The 0.14.x line continues to iterate on the same three stated goals: performance, fee refinement, and decentralization.