TENSORRA/01
MineRunsJobsDocsTreasuryRoadmapAbout

roadmap

Shipped is lit.
Planned is dark.

Dates on shipped items are real commit dates. Nothing below the reticle is promised, only planned. The whitepaper has the launch spec along with the reasoning behind it. This page is the true one.

Read the whitepaper →PDF →

shipped is lit. planned is dark.

  1. 01 / APR 2024

    TENSORRA-01 COMPUTE + WAFER INSTRUMENT

    shipped

    First WebGPU sgemm kernel running in a tab, emitting measured GFLOPS per pass and a wafer scanner bound to live die coverage.

  2. 02 / MAY 2024

    BRAND: TENSORRA LIQUID PALETTE & RETICLE

    shipped

    Obsidian surface, indigo halo, neon settlement accent. The compass reticle became the entry point for every cardinal path.

  3. 03 / JUN 2024

    VERIFIED RUNS, FEED & LEADERBOARDS

    shipped

    Streaming feed of verified runs with GPU model, wallet, TFLOP-hours and settlement hash, ranked per epoch.

  4. 04 / JUL 2024

    TENSORRA-02 MEMORY · TENSORRA-04 MATRIX WORKLOADS

    shipped

    Bandwidth-bound and matrix-bound kernels added so a device is scored on memory and arithmetic, not one synthetic loop.

  5. 05 / AUG 2024

    TENSORRA TREASURY ON SOLANA MAINNET

    shipped

    Vault deployed on mainnet holding tokenized silicon exposure, readable on-chain and mirrored to Solscan.

  6. 06 / SEP 2024

    SIX-HOUR SETTLEMENT + GOVERNOR

    shipped

    Epoch governor closes every six hours, replays sampled passes and releases claimable balances only after attestation.

  7. 07 / OCT 2024

    $TENSORRA FAIR LAUNCH · 0% TEAM ALLOCATION

    shipped

    SPL mint launched with no team allocation, no presale and no reserved supply. Authority revoked at launch.

  8. 08 / NOV 2024

    HOLDER GATE: 15/30/60-MINUTE RUNS

    shipped

    Run duration is unlocked by balance. The gate is checked against the live SPL account before a session starts.

  9. 09 / DEC 2024

    TENSORRA-03 RENDER · TENSORRA-05 INFERENCE ENGINE

    shipped

    Two more workload classes: rasterised render passes and a transformer inference engine compiled to WGSL.

  10. 10 / JAN 2025

    ONLY HOLDERS EARN

    shipped

    Reward eligibility tied to a verified balance for the full duration of a run, snapshotted at both ends.

  11. 11 / FEB 2025

    TREASURY FUNDED WITH REAL-WORLD ASSET TOKENS

    shipped

    Vault holdings moved into tokenized equity exposure across designers, foundries and memory makers.

  12. 12 / MAR 2025

    RENDER JOBS: DISTRIBUTED BROWSER COMPUTE

    shipped

    Job scheduler splits render and batch work across connected tabs, with per-tile proofs recombined at the coordinator.

  13. 13 / APR 2025

    REAL WEIGHTS ON TENSORRA-05

    shipped

    Quantised model weights streamed into the browser runtime, cached per device and checksummed before execution.

  14. 14 / MAY 2025

    INFERENCE API V0

    shipped

    First public endpoint answering completions from the network instead of a hosted cluster.

  15. 15 / JUN 2025

    RECHECKED ANSWERS & VERIFIED COMPUTE ATTESTATION

    shipped

    Sampled responses are recomputed by an independent node and compared before an epoch closes.

  16. 16 / JUL 2025

    HOLDER TIERS & MULTI-GPU MINING

    shipped

    Larger balances mine on more GPUs in parallel, with per-device proofs settled under one wallet.

  17. 17 / AUG 2025

    PUBLIC DATASET OF SUSTAINED BROWSER GPU PERFORMANCE

    shipped

    Open dataset of sustained throughput per adapter, published per epoch from verified runs only.

  18. 18 / PLANNED

    THE MODEL LADDER: 135M TO 1.7B PARAMETERS

    planned

    A ladder of model sizes scheduled by device capability, so a laptop and a workstation both earn on the same network.

  19. 19 / PLANNED

    DEEPSEEK IN A BROWSER TAB

    planned

    Distilled reasoning models compiled to WebGPU, served from consumer hardware with no install.

  20. 20 / PLANNED

    HEADLESS NODES FOR HIGH-VRAM MODELS

    planned

    Optional native runner for operators with large VRAM, sharing the same proof and settlement path.

  21. 21 / PLANNED

    FRONTIER MODELS ON BRING-YOUR-OWN KEY

    planned

    Route frontier requests through a caller-supplied key while network models stay settled on-chain.

  22. 22 / PLANNED

    OPENAI-COMPATIBLE DISTRIBUTED INFERENCE API

    planned

    One wire-compatible endpoint, priced per epoch, answered entirely by verified browser GPUs.