Gemini 3.8 Flash: Intro Pricing Window Through Dec 31, 2026

Coffee Summary

  • Google announced Gemini 3.8 Flash and 3.8 Flash Cyber on September 2, 2026.
  • 3.8 Flash intro price matches 3.7 Flash: $0.75 / $3.75 per 1M input/output tokens through December 31, 2026.
  • From January 1, 2027: $1.50 / $7.50 per 1M input/output.
  • Positioned as the workhorse for coding, agents, and multi-step reasoning; 3.7 Flash remains for efficiency-first workloads.
  • 3.8 Flash Cyber is for trusted defenders via the Fairwind Program only — defender access, not a general API SKU.

What happened

On September 2, 2026, Google introduced Gemini 3.8 Flash (general workhorse) and Gemini 3.8 Flash Cyber (trusted-defender SKU via the new Fairwind Program). Google positions 3.8 Flash as its most intelligent Flash workhorse yet for software engineering, agentic tasks, and multi-step reasoning — at the same introductory price as 3.7 Flash.

Why it matters

For builders, the decision is not only “is 3.8 smarter?” It is “how long does intro pricing last, and when does my unit economics change?”

Google is explicit: intro pricing expires December 31, 2026. On January 1, 2027, list rates step up.

What changed

Pricing window

| Model / period | Input / 1M tokens | Output / 1M tokens |

| — | —: | —: |

| 3.8 Flash intro (through Dec 31, 2026) | $0.75 | $3.75 |

| 3.8 Flash from Jan 1, 2027 | $1.50 | $7.50 |

| 3.7 Flash (efficiency-first; still supported) | Same intro reference as today’s 3.8 intro ($0.75 / $3.75) per Google’s comparison framing | Same |

Net: after the window, 3.8 Flash list price is 2× the intro rates Google published.

Where you can use 3.8 Flash

Per Google’s post:

  • **Developers:** Gemini API via Google AI Studio; agent-first workflows in Google Antigravity; Android Studio; Stitch for UIs.
  • **Enterprises:** Gemini Enterprise.
  • **Consumers:** Google AI Pro and Ultra (Gemini app, AI Mode in Search, Gemini in Sheets).

Google-reported quality signals (not independent)

Cite as Google-reported:

  • Strong showing on **DeepSWE** long-horizon software engineering (Google says it outperforms most larger frontier models on that eval at Flash cost — company claim).
  • **54.9% on HLE-Verified** (Humanity’s Last Exam verified subset), per Google.
  • Gains called out on Vals Finance Agent V2 and Harvey’s Legal Agent Benchmark versus 3.7 Flash / other frontiers (Google-reported).

Google also notes 3.8 Flash may use more tokens on hard tasks (extra reasoning / tool loops), especially at higher effort — so token bills can rise even when per-token price matches 3.7.

3.8 Flash Cyber (brief, defenders only)

Gemini 3.8 Flash Cyber is aimed at trusted defenders through Fairwind (government authorities, critical infrastructure operators, software maintainers — per Google). It focuses on defensive workflows such as vulnerability discovery and automated patching. This article does not cover offensive techniques; if you need Cyber capabilities, apply via Fairwind rather than expecting a public general-availability path.

Who should care

  • Teams standardizing on Flash-class models for coding agents and should plan Q4 2026 vs 2027 budgets now.
  • Cost-sensitive apps that might stay on **3.7 Flash** when efficiency beats peak quality.
  • Security orgs evaluating Fairwind access for Cyber — separately from general 3.8 Flash adoption.

Limitations

  • Benchmarks above are **Google-reported**, not third-party audited here.
  • Post-intro price assumes Google’s published Jan 1, 2027 rates; always re-check the live pricing page before budgeting.
  • Higher diligence can increase token usage — intro parity with 3.7 does not guarantee equal monthly spend.
  • Cyber access is restricted; do not assume API parity with 3.8 Flash.

What to do next

1. If quality matters, switch eval traffic to 3.8 Flash during the intro window and measure task success and tokens/task.

2. Model 2027 cost at $1.50 / $7.50; set an internal go/no-go before Dec 31.

3. Keep 3.7 Flash as the efficiency control lane.

4. Defenders only: review Fairwind eligibility for 3.8 Flash Cyber — skip if you only need general coding/agents.

AIImpish Take

3.8 Flash is Google saying “pay 3.7 prices for a smarter workhorse — until New Year’s.” Use the intro window to measure real tokens-per-success, not just leaderboard screenshots. If your agents get better but burn 2× tokens, January’s rate hike will hurt twice. Lock a migration metric now: quality per dollar through Dec 31, then again under 2027 list prices.