Google launches Gemini 4 Carbon for advanced reasoning and agentic tasks
| Parameter | Before | Now |
|---|---|---|
| Input price | Not specified | $2 per million tokens |
| Output price | Not specified | $10 per million tokens |
What exactly changed in AI models: context, prices, capabilities. Only values stated by sources — never guessed.
| Parameter | Before | Now |
|---|---|---|
| Input price | Not specified | $2 per million tokens |
| Output price | Not specified | $10 per million tokens |
| Parameter | Before | Now |
|---|---|---|
| Task success rate | Not specified | 30% to 100% higher |
| Task completion speed | Not specified | 20% to 30% faster |
| Jailbreak resistance (Argon) | 82% | 99.5% |
| Parameter | Before | Now |
|---|---|---|
| Supported sequence length | Not specified | Up to 14 million tokens |
| Framework version | Not specified | v0.2.4 |
| Parameter | Before | Now |
|---|---|---|
| maxAgentIteration range | 2 to 5 | 0 to 5 |
| Parameter | Before | Now |
|---|---|---|
| Input price | $10 USD per MTok | $8 USD per MTok |
| Output price | $50 USD per MTok | $40 USD per MTok |
| Context window | Not specified | 1M tokens |
| Max output tokens | Not specified | 128k |
| Parameter | Before | Now |
|---|---|---|
| T2I latency (single B200) | 22.36 s | 6.46 s |
| I2I latency (single B200) | 54.64 s | 12.28 s |
| Base T2I latency (8x B200) | 14.90 s | 3.00 s |
| Parameter | Before | Now |
|---|---|---|
| Trainable parameters | Not specified | 16B |
| Parameter | Before | Now |
|---|---|---|
| Capability | Not specified | Vulnerability investigation, reproduction, and fix verification |
| Parameter | Before | Now |
|---|---|---|
| Video duration | 10 seconds | 15 seconds |
| Price | Not specified | $0.07 per second for 720p |
| Parameter | Before | Now |
|---|---|---|
| NVFP4 model footprint reduction (vs FP16) | Not specified | 3.5x |
| NVFP4 model footprint reduction (vs FP8) | Not specified | 1.8x |
| Vera CPU memory capacity | Not specified | 1.5 TB LPDDR5X |
| NVLink-C2C bandwidth | Not specified | 1.8 TB/s |
| Parameter | Before | Now |
|---|---|---|
| Size | Not specified | 65 MB and 75 MB |
| Supported languages | Not specified | English, French, German, Spanish, Portuguese |
| Parameter | Before | Now |
|---|---|---|
| Image Generation Capabilities | Not specified | SVG generation, animated GIF generation, adjustable complexity, markup-guided editing |
| Parameter | Before | Now |
|---|---|---|
| Vera Rubin Architecture | Early developer access | General availability (GA) |
| Parameter | Before | Now |
|---|---|---|
| Parameter count | Not specified | 2 billion |
| Parameter | Before | Now |
|---|---|---|
| Context window | Not specified | 1,048,576 |
| Architecture | Not specified | Mixture-of-Experts (MoE) 2.8T total parameters |
| Vision Encoder | Not specified | MoonViT-V2 |
| Parameter | Before | Now |
|---|---|---|
| Context window | Not specified | 32k tokens (Rerank 5) |
| Supported dimensions | Not specified | 256, 512, 768, 1024, 1536, 2048 |
| Parameter | Before | Now |
|---|---|---|
| Model support | grok-imagine-image-quality and grok-imagine-image-pro | grok-imagine-image-2.0-low |
| Pricing | Not specified | Lower per-image price |
| Parameter | Before | Now |
|---|---|---|
| MCP version | Earlier versions | 2025-11-25 |
| Parameter | Before | Now |
|---|---|---|
| Architecture | Transformer-based models | State Space Models (SSM) |
| First-byte latency | Not specified | Approx 90ms |
| Parameter | Before | Now |
|---|---|---|
| Request structure | Full conversation history sent each time | Only new messages sent via thread |
| Time to first token | Standard | Reduced |
| Parameter | Before | Now |
|---|---|---|
| Cosmos3-Nano Generation (VRAM needed in BF16) | Not specified | 58 GiB |
| Cosmos3-Nano Generation (VRAM needed in FP8) | Not specified | 44 GiB |
| Cosmos3-Nano Generation (VRAM needed in FP8 + offloading) | Not specified | 31 GiB |
| Parameter | Before | Now |
|---|---|---|
| OSWorld-Verified benchmark score | Not specified | 66.0% |
| Parameter | Before | Now |
|---|---|---|
| Dynamo platform version | Not specified | 1.4.2 |
| DynamoGraphDeployment CRD | v1alpha1 | v1beta1 |
| NATS dependency | Mandatory | Optional opt-in |
| Request plane protocol | Not specified | TCP |
| KV-cache event protocol | Not specified | ZMQ |
| Parameter | Before | Now |
|---|---|---|
| Fast mode | Available | Deprecated/Removed |