GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

Google Ships Three Gemini Models: 3.6 Flash at $1.50/M, 49% DeepSWE, and a Restricted Cyber Tier

Google shipped three Gemini models on July 21, spanning efficiency, cost, and a restricted cybersecurity tier that the company is positioning as a frontier competitor to Anthropic’s Fable 5.

Gemini 3.6 Flash: The Workhorse Gets Cheaper and Better

Gemini 3.6 Flash is the headliner. Google describes it as a drop-in improvement over 3.5 Flash in every dimension that matters for production agentic workloads.

Benchmark gains over 3.5 Flash:

Benchmark3.5 Flash3.6 Flash
DeepSWE37%49%
MLE Bench49.7%63.9%
OSWorld-Verified78.4%83.0%

The efficiency number is notable: 3.6 Flash generates 17% fewer output tokens than 3.5 Flash on the Artificial Analysis Index, and in coding-heavy workloads like DeepSWE, token reduction reaches up to 65% in some configurations. Fewer tokens means fewer reasoning loops and tool calls per task.

Pricing sits at $1.50/1M input, $7.50/1M output — lower than 3.5 Flash per output token. For developers running high-volume agentic workflows, the effective cost-per-task drops further than the headline price suggests because the model simply writes less.

Computer use is now a built-in client-side tool rather than a separate API call.

Gemini 3.5 Flash-Lite: 350 Tokens Per Second

The Flash-Lite tier targets the fastest, cheapest end of the range. Google says 3.5 Flash-Lite delivers 350 output tokens per second on the Artificial Analysis Index, and “significantly outperforms” prior Flash-Lite generations on agentic workflows. No benchmark numbers were published for the model itself, but at 350 tok/s it is competitive with Mercury 2 and LFM2.5-VL-1.6B, which currently hold the top output speed positions on Artificial Analysis.

Gemini 3.5 Flash Cyber: Government-Only, CodeMender Bundled

The third model is a departure from the Flash naming convention in substance if not label. Gemini 3.5 Flash Cyber is a specialized, highly efficient model for finding and fixing software vulnerabilities. It ships paired with Google’s CodeMender code security agent — a deliberate product bundle rather than a raw model release.

Access is restricted to governments and trusted partners at launch. CNBC described it as Google’s “clearest answer yet to Anthropic’s lead in cybersecurity,” a reference to Fable 5/Mythos, which also launched under government-restricted access before wider rollout.

Google’s framing is significant: instead of positioning Flash Cyber as a raw model, they shipped a complete agent product. CodeMender handles orchestration. Flash Cyber handles the low-level vulnerability analysis. Google claims competitive performance at the frontier tier.

What Is Not Here: Gemini 3.5 Pro

Despite analyst expectations for a Gemini 3.5 Pro announcement, Google confirmed it is still in partner testing. The blog post says it will be “broadly available as soon as it’s ready.” That language does not suggest imminent release.

What Is Coming: Gemini 4

Buried at the bottom of the announcement: Google has started its “most ambitious pre-training run yet” for Gemini 4. No timeline, no capability claims, no benchmark targets. It is the first public confirmation that Gemini 4 pre-training is active.

Context

The three-model drop positions Google to compete across cost tiers while reserving its highest-capability, safety-restricted product for institutional buyers. Gemini 3.5 Pro and Gemini 4 are the two known items still in the pipeline. Until Gemini 3.5 Pro arrives, the competitive gap between Google’s publicly available models and Anthropic’s Fable 5 on the Artificial Analysis Intelligence Index remains roughly 15 points.