
Gemini 4 vs 3.5: Flash, the Missing Pro, Price & Access
By Nathan Cole
MyClaw Editorial
Gemini 4 Argon
Try Gemini 4 Argon
Start a chat with Gemini 4 Argon for coding, writing, or research.
Chat with Gemini 4 ArgonAI Takeaway
- Does Gemini 4 beat Gemini 3.5? On Google's own numbers, Gemini 4 (Argon) is far ahead on long software tasks. It scores 77.9 on DeepSWE v1.1, against 37 for Gemini 3.5 Flash. The scores come from separate Google tables published months apart, so treat the gap as a direction, not an exact margin.
- What happened to Gemini 3.5 Pro? Google announced it at I/O in May for a June release, then kept it in partner testing all summer. On September 30, a Google spokesperson told Reuters the company no longer plans to release it. So any Gemini 4 vs Gemini 3.5 Pro comparison sets Gemini 4 against a model that never shipped.
- Which costs more, Gemini 4 or Gemini 3.5 Flash? Gemini 4, slightly. Google's introductory rate is $2/$10 per 1M input/output tokens, rising to $4/$20 later. Gemini 3.5 Flash is $1.50/$9.00 on Google's price list.
- Can you use Gemini 4 today? Most people can't yet. Access comes in waves, beginning with Google's Fairwind Program, and no date has been set for the next groups. Gemini 3.5 Flash has been generally available in the Gemini API since May.
- Can you test Gemini 4 against Gemini 3.5 on MyClaw? You can. MyClaw offers both Gemini 4 and Gemini 3.5 Flash, so one hosted agent can run the same recurring job on each model.
Gemini 4 vs 3.5: The Spec Sheet
The Gemini 4 vs 3.5 matchup puts Google's September 2026 flagship next to the Flash model that opened the 3.5 generation in May.
| Spec | Gemini 4 (Argon) | Gemini 3.5 Flash |
|---|---|---|
| API model ID | None public yet | gemini-3.5-flash |
| Status | Staged rollout, announced Sep 30, 2026 | Generally available since May 19, 2026; Google now lists it as legacy |
| Context window | Not published | 1M |
| Output limit | Up to 1M tokens | 64K tokens |
| Inputs | Google highlights charts, long video and documents | Text, images, audio and video |
| Thinking | Benchmarked at Google's highest setting; levels not listed | Minimal, low, medium (default) and high |
| Computer use | Not stated | Not supported |
Output length is the clearest jump. Gemini 3.5 Flash stops at 64K tokens per reply. Gemini 4 can return as many as 1M, which helps with long reports, full code files and large data rewrites. Our Gemini 4 Argon review covers what Google has shared so far.

Which Gemini 3.5 Is Gemini 4 Replacing?
"Gemini 3.5" covers four text models, and only some of them reached developers:
- 3.5 Flash launched May 19 and was Google's default Flash model until newer releases took over.
- 3.5 Flash-Lite arrived July 21 as the cheap, fast tier.
- 3.5 Flash Cyber is 3.5 Flash fine-tuned to find and fix security flaws, limited to government users through Google's CodeMender.
- 3.5 Pro was shown at I/O, tested with partners, and then dropped.
That settles the Gemini 4 vs Gemini 3.5 Pro question: there's no 3.5 Pro to benchmark. Gemini 4 takes the slot 3.5 Pro was meant to fill. Reuters reports that Argon is larger than Google's earlier Pro models. The Pro-tier Gemini you can buy today is still Gemini 3.1 Pro, which we tested against 3.5 Flash in our 3.1 Pro and 3.5 Flash agent test.
So this article compares Gemini 4 with 3.5 Flash, the 3.5 model most teams actually used. Google has since shipped 3.6, 3.7 and 3.8 Flash, so 3.5 Flash isn't the newest Flash either. For the budget tier, see how Flash-Lite stacks up against 3.5 Flash. Speech models with 3.5 in the name, such as Live Translate, aren't covered here.

Benchmarks: Two Google Tables, Four Months Apart
Google hasn't published a single Gemini 4 vs 3.5 table. Two benchmarks do appear in both its September Gemini 4 table and its earlier 3.5 Flash results:
| Test (Google figures) | Gemini 4 (Sep 2026) | Gemini 3.5 Flash |
|---|---|---|
| DeepSWE v1.1 (long software tasks) | 77.9 | 37 (July table) |
| Finance Agent v2 (Vals) | 65.4 | 57.9 (May model card) |
The Gemini 4 vs Gemini 3.5 Flash gap is huge on DeepSWE, which tests long, multi-step engineering work. It's much smaller on finance agent tasks. Treat both rows as directional: Google ran Gemini 4 at its highest thinking setting, computed several scores itself, and published the 3.5 Flash numbers in May and July.
The other rows don't line up. Gemini 4's table reports 51.3 on AutomationBench, 57.4 on Terminal-Bench 4.0 and 84.2 on GraphWalks (256K–1M tokens). The 3.5 Flash card reports Terminal-Bench 2.1 (76.2%), MCP Atlas (83.6%) and CharXiv Reasoning (84.2%). The two Terminal-Bench versions are different tests, so 57.4 against 76.2 doesn't mean Gemini 4 got worse. Google's full Gemini 4 table, rival models included, is in our Fable 5.1 comparison.
Gemini 4 vs 3.5 Pricing: Closer Than You'd Expect
| Price per 1M tokens (Google, standard) | Input | Output |
|---|---|---|
| Gemini 4, introductory | $2.00 | $10.00 |
| Gemini 4, after the introductory period | $4.00 | $20.00 |
| Gemini 3.5 Flash | $1.50 | $9.00 |
| Gemini 3.5 Flash-Lite | $0.30 | $2.50 |
| Gemini 3.8 Flash, through Dec 31, 2026 | $0.75 | $3.75 |
| Gemini 3.8 Flash, from Jan 1, 2027 | $1.50 | $7.50 |
Picture an agent call with 200K tokens of input, 20K of output and no caching. At Google's rates, it costs about $0.48 on Gemini 3.5 Flash and $0.60 on Gemini 4 at launch, roughly 25% more. After Gemini 4's launch pricing ends, the same call costs about $1.20, or 2.5 times the 3.5 Flash bill. Google hasn't said when that happens.
Other Flash models cost less: about $0.23 on 3.8 Flash this year and $0.11 on 3.5 Flash-Lite. If price per call decides it, a newer Flash may beat both sides of this Gemini 4 vs 3.5 comparison. Our Gemini 3.8 Flash review covers that option. Token use per task also varies by model.
Who Can Use Each Model Today
Gemini 4 is reaching people in stages. Cybersecurity partners in Google's Fairwind Program get it first. Google AI Ultra subscribers and paying API customers come next, but Google hasn't named a date. There's no public model ID either, so most developers can't call it yet.
By contrast, Gemini 3.5 Flash is generally available in the Gemini API through Google AI Studio, in Google Antigravity and in Gemini Enterprise Agent Platform, at stable list prices. But Google's model list now labels it legacy, and newer Flash models have replaced it as the default.
Gemini 4 vs 3.5 Verdict: Upgrade Now or Wait?
For the Gemini 4 vs Gemini 3.5 Flash choice, pick Gemini 4 for long, hard agent and coding jobs and very long outputs, where output quality outweighs cost. You'll need access first. Stay on Gemini 3.5 Flash, or move to a newer Flash, for high-volume work you need to run today.
Gemini 4
Pros:
- Google's newest flagship
- Large Google-reported lead on DeepSWE v1.1
- Replies of up to 1M tokens
- Launch price close to 3.5 Flash
Cons:
- Staged access with no date for most users
- Price doubles after the introductory period
- No public model ID or context window
- Only Google's own benchmarks so far
Gemini 3.5 Flash
Pros:
- Generally available and stable
- $1.50/$9.00 per 1M tokens (Google)
- 1M-token context window
- Four thinking levels, including minimal and low
Cons:
- Now a legacy model, with cheaper newer Flash options
- 64K output cap
- No computer use
- Far behind Gemini 4 on DeepSWE v1.1
Run Gemini 4 and Gemini 3.5 Flash on One Agent Job

Two separate Google tables can't show which model handles your work better. On MyClaw, both models can run in the same hosted OpenClaw or Hermes Agent. Give one recurring job, such as inbox triage, a repo task or a weekly report, to each side by side, and settle Gemini 4 vs 3.5 with real output. Then compare quality, latency, retries and spend. Keep Gemini 4 where the extra quality pays off and 3.5 Flash for volume. For more options, see our OpenClaw model picks.
Gemini 4 vs 3.5 on MyClaw: Three Steps
- Start an agent. Log in to MyClaw and set up a new OpenClaw or Hermes Agent.
- Choose the models. Turn on Gemini 4 and Gemini 3.5 Flash. Connect your chat app too.
- Compare on live work. Hand both models the same weekly task, then rate the answers for accuracy, speed and retries.
Put Gemini 4 Argon to work on a real task
Start a chat with Gemini 4 Argon for coding, writing, or research.
Chat with Gemini 4 Argon