TRACKING LIVE

GPT-5.5 "Spud" Release Tracker

Real-time countdown, leaked benchmarks, pricing predictions, and frontier model comparison for OpenAI's next model.

Estimated Release Window

Based on pre-training completion (late March 2026) + typical post-training timeline

--
Days
--
Hours
--
Minutes
--
Seconds

Earliest estimated date: May 15, 2026 | Latest: June 30, 2026

Countdown shows time to earliest estimate. Actual date may be later.

Development Timeline

Q3 2024
Research Phase Begins
OpenAI starts two-year research cycle incorporating lessons from GPT-5 and GPT-5.4.
January 2026
Pre-training Starts
Large-scale pre-training begins on OpenAI's expanded compute cluster. Codename "Spud" first appears in internal communications.
Late March 2026
Pre-training Complete
Pre-training finishes. Sam Altman and Greg Brockman make public comments describing "a strong new base model." Post-training alignment and RLHF begins.
April 2026 (NOW)
Internal Testing & Red-teaming
Model undergoing safety evaluation, red-teaming, and internal benchmarking. Select partners may have early access.
May–June 2026
Expected Public Release
API access for developers, ChatGPT Plus/Pro integration, and public benchmarks. Pricing announcement expected simultaneously.
Q3 2026
GPT-5.5 Mini Expected
Smaller, cheaper variant following OpenAI's pattern of releasing mini versions 2-3 months after the full model.

🔮 Capability Predictor

Estimate how GPT-5.5 will perform on your use case based on leaked signals and benchmark trends.

GPT-5.5 Spud Impact Prediction

What We Know So Far

Pre-training Complete

Sam Altman confirmed "a strong new base model" has finished pre-training, incorporating two years of research advances.

HIGH CONFIDENCE
Source: Sam Altman public statements, The Information, March 2026

Native Multimodality

Unlike GPT-5.4's modular approach, Spud is reported to have unified text/image/audio/video understanding in a single model architecture.

HIGH CONFIDENCE
Source: OpenAI job postings, developer preview hints, multiple leakers

Enhanced Agentic Capabilities

Significant improvements in multi-step task execution, tool chaining, and autonomous workflow completion. Built-in computer use abilities.

HIGH CONFIDENCE
Source: OpenAI Agents SDK updates, Greg Brockman comments

Codename "Spud"

Internal codename following OpenAI's tradition of food-related names. Some speculate it references the model's "growth from a seed" architecture.

MEDIUM CONFIDENCE
Source: Internal communications, The Verge, leaker accounts

1M+ Context Window

Expected to support 1 million+ tokens in context, competing with Gemini's long-context capabilities.

MEDIUM CONFIDENCE
Source: Developer preview speculation, competitive pressure analysis

Pricing: $10-20/M Input

Estimated at $10-20 per million input tokens and $30-60 per million output tokens. Mini variant expected at 5-10x cheaper.

SPECULATIVE
Source: Pricing trend analysis, competitive positioning

GPT-5.5 Spud vs Frontier Models (Predicted)

Benchmark GPT-5.5 Spud (Est.) GPT-5.4 Claude Opus 4.7 Gemini 2.5 Ultra Llama 4 Behemoth
MMLU-Pro ~94.5% 91.2% 92.8% 93.1% 90.5%
HumanEval+ ~96% 92.4% 94.1% 91.8% 89.2%
MATH-500 ~98% 94.8% 96.2% 95.5% 92.1%
GPQA Diamond ~78% 71.4% 74.9% 73.2% 69.8%
SWE-bench Verified ~72% 65.3% 68.7% 62.1% 58.4%
Agentic Tasks ~88% 79.5% 85.2% 80.1% 75.3%
Context Length 1M+ tokens 128K 200K 2M 512K
Price (Input/1M) ~$15 $10 $15 $12.50 Open source

*GPT-5.5 Spud scores are estimates based on leak analysis and trend extrapolation. Actual results may differ.

Frequently Asked Questions

What is GPT-5.5 Spud?▼
GPT-5.5 "Spud" is OpenAI's upcoming next-generation language model. Pre-training reportedly finished in late March 2026. The codename "Spud" follows OpenAI's tradition of food-related internal codenames. It incorporates two years of research, native multimodality, and enhanced agentic capabilities, making it a significant leap over GPT-5.4.
When will GPT-5.5 Spud be released?▼
Based on leaks and statements from OpenAI leadership, GPT-5.5 Spud is expected to launch between late April and June 2026. Pre-training completed in late March 2026, and internal testing is underway. The exact date has not been officially announced. A mini variant is expected 2-3 months after the full release.
How will GPT-5.5 compare to Claude Opus and Gemini?▼
GPT-5.5 Spud is expected to significantly advance math, coding, and agentic capabilities over GPT-5.4. Based on leaked information, it may match or exceed Claude Opus 4.7 in reasoning benchmarks and compete with Gemini 2.5 Ultra in multimodal tasks. Key differentiators will likely be native tool use, improved instruction following, and enhanced agentic workflow capabilities.
How much will GPT-5.5 cost?▼
Pricing has not been announced. Based on GPT-5.4 pricing trends and competitive pressure from Claude and Gemini, estimates suggest $10-20 per million input tokens and $30-60 per million output tokens for the full model, with a cheaper "mini" variant likely following within months at 5-10x lower cost.
What new features will GPT-5.5 Spud have?▼
Expected new features include: native multimodal input/output (text, image, audio, video in a single architecture), enhanced agentic capabilities (multi-step task execution, tool chaining, computer use), improved math and scientific reasoning, better code generation and debugging, longer context windows (potentially 1M+ tokens), and native function calling improvements.
Should I wait for GPT-5.5 or use GPT-5.4 now?▼
If your workload is well-served by GPT-5.4, there's no need to wait. GPT-5.5 will likely offer meaningful improvements in reasoning, agentic tasks, and multimodal processing, but GPT-5.4 remains highly capable. For cost-sensitive applications, consider that GPT-5.5 may be priced higher initially. Plan to evaluate and migrate when the model is available.