Gemini 3.6 Flash: 17% Fewer Tokens, Live in Copilot

Google shipped Gemini 3.6 Flash today, July 21, 2026, and the timing matters. The flagship Gemini 3.5 Pro — announced at Google I/O 2026 in May — is still stuck in testing after falling short on coding performance and reliability. Flash, meanwhile, is here: cheaper than its predecessor, more efficient, and available in GitHub Copilot for Pro, Pro+, Max, Business, and Enterprise users on day one. If you’re running agentic workflows on 3.5 Flash, this is a straightforward upgrade.

Gemini 3.6 Flash: Performance and Pricing Changes

Gemini 3.6 Flash uses 17% fewer output tokens than 3.5 Flash on the Artificial Analysis Index — and in some DeepSWE coding tasks, the reduction reached 65%. Output pricing dropped from $9/1M to $7.50/1M tokens. Input remains at $1.50/1M.

That 17% average isn’t marketing copy. At 10 million output tokens per day, you move from $90/day to $75/day — $450/month saved with a one-line model string change. Agents doing 100M tokens/month recover $4,500/month. The savings compound further because 3.6 Flash also completes more tasks on the first attempt, wasting fewer tokens on retries and restarts.

On benchmarks that matter for agentic work: DeepSWE (autonomous software engineering) jumps from 37% to 49%; OSWorld Verified (computer-use tasks) climbs from 78.4% to 83%; MLE Bench (ML engineering) goes from 49.7% to 63.9%. The knowledge cutoff also advances from January 2025 to March 2026 — a 14-month jump that meaningfully reduces stale-knowledge hallucinations for many workloads.

Related: LLM API Costs Dropped 94%: What to Fix in Your Architecture Now

GitHub Copilot: Available Today, But Read the Fine Print

Gemini 3.6 Flash is in GitHub Copilot’s model picker as of today across VS Code, JetBrains, Xcode, Eclipse, and the Copilot CLI. Individual-tier users (Pro, Pro+, Max) can select it immediately from the model dropdown. No API setup required.

Enterprise and Business plan users face one extra step: an administrator must enable the “Gemini 3.6 Flash Preview” policy in Copilot settings before anyone in the org can access it. Also worth knowing — Copilot billing for this model moves to usage-based pricing at provider list rates, not the flat-rate subscription. Budget accordingly before enabling it org-wide.

Related: Kimi K2.7 Code Lands in GitHub Copilot: Open-Weight, Finally

One Breaking Change Before You Migrate

The Gemini API changelog for 3.6 Flash deprecates the temperature, top_p, and top_k sampling parameters. If your integration passes these, the API will reject the request after you switch model IDs. Fix it first:

# Before (3.5 Flash) — breaks on 3.6 Flash
response = client.models.generate_content(
    model="gemini-3.5-flash",
    contents=prompt,
    config={"temperature": 0.7, "top_p": 0.9}
)

# After (3.6 Flash) — remove deprecated sampling params
response = client.models.generate_content(
    model="gemini-3.6-flash",
    contents=prompt
)

The parameter removal is intentional — 3.6 Flash uses configurable reasoning effort instead of raw sampling controls. Review the changelog for migration guidance before touching production.

Pro Is Still Delayed. Gemini 4 Is Next.

3.6 Flash isn’t a routine model bump — it’s Google keeping developer momentum alive while the flagship gets rebuilt. Gemini 3.5 Pro was supposed to ship in June 2026 after its I/O announcement. Internal testing found coding performance, hallucinations, and multi-step reasoning all below the bar set at the keynote. It’s still in testing as of today, with no hard ship date.

In the same blog post announcing 3.6 Flash, Google teased Gemini 4 as “the most ambitious pre-training run yet” — no specs, no timeline, just a signal that the roadmap hasn’t stopped. The clear implication: Flash is doing the work right now, Pro is on hold, and the next major bet is Gemini 4. If you’re planning infrastructure around a Pro-tier Gemini model, that timeline is still genuinely unclear.

Key Takeaways

• Gemini 3.6 Flash ships today with 17% fewer output tokens and lower pricing ($7.50/1M output vs $9/1M) — a free efficiency upgrade for anyone on 3.5 Flash.
• GitHub Copilot integration is live across all paid tiers; Enterprise and Business admins must enable the Preview policy and account for usage-based billing.
• Before migrating, remove temperature, top_p, and top_k from your API calls — these parameters are now deprecated and will break requests on 3.6 Flash.
• Gemini 3.5 Pro is still delayed with no firm ship date; Flash is carrying Google’s developer lineup for now.
• Gemini 4 was teased in today’s announcement — described as “the most ambitious pre-training run yet,” with no timeline given.