Gemini 3.6 Flash Ships — 3.5 Pro Still Waiting

On July 21, Google made Gemini 3.6 Flash and Gemini 3.5 Flash-Lite generally available on the Gemini API — and still did not ship Gemini 3.5 Pro. The official blog frames 3.6 Flash as the new workhorse for coding, knowledge work, and multimodal agents, with lower output verbosity than 3.5 Flash. Pricing lands at $1.50 input and $7.50 output per million tokens. Logan Kilpatrick said 3.5 Pro is testing with partners and will land when ready; Gemini 4 pre-training has already started. For teams waiting since Google's May I/O tease of a June Pro launch, that is another week of Flash-only news.

helloai still tracks Gemini 3.1 Pro Preview as the Google frontier entry: $2/$12 for prompts under 200K tokens, 1M context, and 1486 Elo on LMArena text overall as of July 21. That is deliberate. Admission rules keep one model per positioning slot, and Flash does not replace Pro on hard reasoning even when arena scores get close. Gemini 3.6 Flash already sits at 1485 Elo with preliminary votes — one point under 3.1 Pro and inside the 25-point floor of the tracked set. It would pass the Elo hard filter if scored as a new Google flagship. It fails the product test: same provider, speed-and-cost tier, and no claim on the Hard Reasoning & Science niche 3.1 Pro still holds.

The delay is no longer rumor. Bloomberg reported mid-July that 3.5 Pro missed internal performance goals; TechCrunch and Ars covered the July 21 Flash release as confirmation that Pro remains gated. Google's own post says 3.5 Pro will be broadly available "as soon as it's ready." Meanwhile rivals moved: Claude Fable 5 leads helloai's set at 1507 Elo, Muse Spark 1.1 sits second at 1495 after this week's arena refresh, and Moonshot's Kimi K3 holds 1486 on the public API. Google's Flash line is getting more efficient; its Pro line is getting older relative to the board.

What developers should do this week is separate agent economics from flagship aspiration. If your workload is multi-step tool use and long-horizon planning where output tokens dominate cost, retest 3.6 Flash first — the rate card is live on Google's pricing page. If you need PhD-level science, multimodal understanding, or the highest-capability Google brain you can buy today, 3.1 Pro remains the helloai pick until 3.5 Pro appears on that same pricing page with public API access.

helloai will not promote 3.6 Flash into the six-model frontier table while 3.1 Pro is still the documented multimodal leader and 3.5 Pro is still not GA. The moment 3.5 Pro ships with transparent rates and arena signal, it becomes a same-provider replace candidate for gemini, not an expansion. Until then, the news is efficiency, not a new crown.