Inkling ships 975B open-weights model on Product Hunt

Inkling shipped a 975-billion-parameter open-weights multimodal model on Product Hunt's July 20 daily leaderboard, picking up 192 votes, per the listing.

For teams whose architecture assumes a rented API, a fine-tunable base at this scale turns self-hosting from a lab-only exercise into a live option.

The model is built for fine-tuning, not for hosted inference, per the listing. That's a different bet than the usual frontier release, aimed at people who want a base to adapt rather than an endpoint to call.

Frontier-scale open-weights releases have mostly surfaced through lab blog posts or arXiv preprints, aimed at researchers comparing benchmarks. This one landed on a consumer product board instead, ranked next to whatever else shipped that day.

Teams evaluating a self-hosted base have another candidate to test, whatever the vote count says about quality. The listing doesn't say how those 192 votes translate to benchmark results, or whether anyone's tested it against hosted frontier models.