THE DISTANCE BETWEEN
Every promise leaves a question.
We keep the question open until
the evidence catches up.
THREE THREADS TO FOLLOW
Omri described Pearl-GEMM as in testnet on September 25. Mainnet activation is not verified in this record.
The proof-of-inference app was described as an L2/rollup. That statement establishes a direction, not deployment.
The September AMA reported provider conversations and unnamed neolabs mining Pearl. The counterparties remain unidentified.
Historical record through September 17, with selected public-source checks on September 21 and protocol/L2 statements reviewed September 25. Each entry identifies its evidence date. Publication, announcement and activation remain separate states.
THE RECORD / ENTRY BY ENTRY
done
Together announced its Pearl endpoint in May. On September 21 its model page reported Serverless unavailability. Other deployment options and the transition date were not verified.
Inspect the source ↗Source: Public source reviewed; scope stated in claim. Checked: 2026-09-21.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
2026-09-22 · Research edition: separated observed evidence from deployment and adoption inferences.
Jul 2026
The MoE hard fork shipped, then MoE mining was disabled following the July vulnerability. The supplied post-mortem says it remained off pending a revised construction and review; later reactivation is not established here.
Inspect the source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
2026-07-23 · Disabling soft fork described in the July post-mortem recorded on the timeline.
Sep 14 2026
FP8 research and code were published for review. PR #311 opened September 9; September 14 is the announcement date in the archive. Performance remains team-reported.
Inspect the source ↗Source: Public source reviewed; scope stated in claim. Checked: 2026-09-21.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
2026-09-22 · Research edition: separated observed evidence from deployment and adoption inferences.
Sep 14 2026
Partnership with attestable — ZK proofs × PoUW as a "trust layer for AI"
Inspect the source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
Sep 25: expected within weeks
On September 25, Omri described the new Pearl-GEMM as already in testnet and expected to launch within weeks. This updates the earlier publication record; mainnet activation remains unverified.
Inspect the September 25 source ↗Source: Public source reviewed; scope stated in claim. Checked: 2026-09-21.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
2026-09-22 · Research edition: separated observed evidence from deployment and adoption inferences.
2026-09-25 · Original public X reply inspected; team statement verified, implementation and performance not independently tested.
Historical estimate; no confirmed delivery date
FP4 scheme — the "endgame," added alongside FP8; "a couple more months of work"
Inspect the source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
WIP
The September AMA describes unfinished Hopper support for the proposed FP8 scheme. This does not imply that the existing integer miner lacks Hopper support.
Inspect the source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
teased
The September AMA describes an institutional-access ambition without naming a product or counterparty. Anchorage and Coinbase remain hostname-based editorial theories.
Inspect the source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
open
"Two other major inference providers" in touch + "several neolabs already mining Pearl" — none named
Inspect the source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
About one month after the earlier call
Chip-maker collaboration goes public (a joint reveal) — still unannounced as of the Sept AMA
Inspect the source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
near term
Pearl advertises serverless API, managed inference and on-premise services. Successful access and fulfillment of the earlier marketplace promise were not independently verified.
Inspect the source ↗Source: Public source reviewed; scope stated in claim. Checked: 2026-09-21.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
2026-09-22 · Research edition: separated observed evidence from deployment and adoption inferences.
Sep 25 · no launch date
The September AMA discussed L2 scaling research. On September 25, Omri explicitly placed the proof-of-inference app on an L2/rollup, outside L1 Pearl-GEMM. The reply does not establish deployment, a Simplex implementation or the payment architecture.
Inspect the September 25 source ↗Source: Supplied editorial record. Checked: historical archive; no new verification.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
2026-09-25 · Original public X reply inspected; team statement verified, implementation and performance not independently tested.
Long-term aspiration; not a committed release date
Pearl documents a vLLM mining plugin. Native inclusion in upstream vLLM remains unverified.
Inspect the source ↗Source: Public source reviewed; scope stated in claim. Checked: 2026-09-21.
2026-09-20 · Migrated the supplied historical record into explicit lifecycle states; not a fresh external verification.
2026-09-22 · Research edition: separated observed evidence from deployment and adoption inferences.
Read it against what's already pinned on this board, and the roadmap supplies three questions to keep investigating:
VERBATIM FROM THE PEARL COMMUNITY DISCORD, PUBLIC CHANNEL. THE READ-ACROSS IS OURS.
Archived community discussion: leads and arguments to investigate. The supplied record preserves these quotations with identities withheld; they are not independently confirmed claims.
Members spotted Pearl featured in a Google Cloud (GCP) banner. The prevailing read in the channel: it's the standard GCP startup-program tie — "partners like many other startups" — not a signed serving deal.
Odds it's a major GCP deal: low. But it's a fresh pin on the board.
The doubters' cut: nobody outside Together AI is racing to serve Pearl-ified open models. If a 2-for-1 Gemma were meaningfully better, the argument goes, rival platforms would be scrambling for it — and so far they're not.
The bull reply: that gap is exactly what a mining subsidy exists to close — which is why the next hyperscaler signature matters so much.
The recurring wish in the channels: the next hot open model — GLM 5.2 and friends — Pearl-ified and served somewhere beyond a single Gemma build on Together AI.
Fireworks is the name that keeps coming up. Reinforces the top hyperscaler suspect.