TokenMix Research Lab · 2026-04-17

GPT-5.5 (Spud) Released: $5/$30 API Pricing & Benchmarks 2026

Last Updated: 2026-05-14 Author: TokenMix Research Lab

OpenAI shipped GPT-5.5 on April 23, 2026 — codename Spud confirmed in the official launch post. The model is live in the API at gpt-5.5, priced at $5 input / $30 output per million tokens, with a 1M-token context window. GPT-5.5 Pro is also live at $30 / $180. Terminal-Bench 2.0 jumped from 75.1% (GPT-5.4) to 82.7% in a single release. GPT-5.5 Instant followed for free-tier users on May 5, 2026. Below is everything confirmed at launch, the real benchmark deltas against Claude Opus 4.7 and Gemini 3 Pro, and how to call gpt-5.5 today through TokenMix.ai with no waitlist.

Table of Contents

• Quick Verdict: GPT-5.5 Launch Snapshot
• What Actually Shipped on April 23, 2026
• GPT-5.5 API Pricing: All Tiers, Batch & Flex Discounts
• Real Benchmark Results vs GPT-5.4, Claude Opus 4.7, Gemini 3 Pro
• Cost Per Task: GPT-5.5 vs Competitors on Realistic Workloads
• How to Access GPT-5.5 API Today (Direct vs TokenMix)
• Migration Checklist: GPT-5.4 → GPT-5.5
• FAQ

Quick Verdict: GPT-5.5 Launch Snapshot

Bottom line: GPT-5.5 is the new agentic-coding leader (Terminal-Bench 82.7%) and dominates long-context retrieval at 512K-1M tokens (74.0% vs Opus 4.6 32.2%). It is not the across-the-board winner — Claude Opus 4.7 still leads on SWE-Bench Pro, and Gemini 3 Pro leads on ARC-AGI-1.

What Actually Shipped on April 23, 2026

Spud is no longer speculation. As confirmed in OpenAI's launch post, GPT-5.5 is now generally available across three surfaces:

ChatGPT: GPT-5.5 Thinking rolled out to Plus, Pro, Business and Enterprise users on April 23. GPT-5.5 Pro went to Pro, Business and Enterprise the same day. GPT-5.5 Instant — a free-tier variant — followed on May 5, 2026.

Codex: GPT-5.5 is available on Plus, Pro, Business, Enterprise, Edu and Go plans, with a 400K context window. A "Fast" mode generates 1.5× faster at 2.5× the cost.

API: gpt-5.5 and gpt-5.5-pro are exposed through Responses and Chat Completions endpoints. Microsoft Foundry has parity access. Both models support the full 1M-token context window via the API.

The "Spud" codename was used internally during training and publicly acknowledged after release. The model uses NVIDIA GB200 and GB300 NVL72 systems for serving. Per-token latency matches GPT-5.4 despite the intelligence jump — OpenAI explicitly co-designed inference infrastructure during training. According to TokenMix.ai upstream uptime tracking, gpt-5.5 has been served stably since the day-of-launch ramp.

GPT-5.5 API Pricing: All Tiers, Batch & Flex Discounts

Per OpenAI's pricing page referenced in the launch post, GPT-5.5 is twice the price of GPT-5.4 on inputs and twice on outputs. OpenAI's argument is that GPT-5.5 uses fewer tokens per task in Codex, partially offsetting the higher per-token rate.

Standard API Pricing

Note: OpenAI's launch post labels the Gemini comparison column "Gemini 3 Pro". Google's own pricing page calls the model "Gemini 3 Pro" with API ID gemini-3.1-pro-preview. They are the same model.

Modal Pricing Variants (GPT-5.5)

Translation: if you can tolerate 24-hour completion, Batch puts GPT-5.5 at $2.50 / $15 per MTok — exactly matching old GPT-5.4 standard pricing. According to TokenMix.ai pricing observability, this is the sweet spot for batch summarization or large-scale classification workloads.

Real Benchmark Results vs GPT-5.4, Claude Opus 4.7, Gemini 3 Pro

These are not projections. All numbers below are from OpenAI's launch announcement, which published third-party benchmark scores at release.

Agentic Coding (GPT-5.5's Strongest Domain)

*SWE-Bench Pro Public has known memorization issues — OpenAI flagged this in the launch post. Treat with caution.

Knowledge Work & Tool Use

Math & Science

Long Context (Where GPT-5.5 Pulls Decisively Ahead)

The 1M-token reasoning result is the most striking single delta in this launch. GPT-5.4 collapsed to 9.4% on Graphwalks BFS at 1M context. GPT-5.5 holds 45.4%. For RAG over very large corpora or full-codebase analysis, this is the first model that does not require manual chunking.

Cost Per Task: GPT-5.5 vs Competitors on Realistic Workloads

Per-token pricing alone is misleading because token efficiency varies. OpenAI claims GPT-5.5 uses fewer tokens per task than GPT-5.4. Below are costs for three common workflows, computed from public pricing as of 2026-05-14.

*Claude Opus 4.7 capped at 200K context. Workloads requiring >200K input must split or use GPT-5.5 / Gemini 3 Pro. **Gemini 3 Pro at 200B in / 50B out hits the >200K tier: 200B × $4 + 50B × $18 = $800K + $900K = $1.7M; with Batch 50% off = $850K. Numbers reflect estimated mix of ≤200K and >200K prompts. ***Anthropic publishes Batch processing in their docs left-nav but tier discount specifics not re-verified here.

Three observations:

• Claude Opus 4.7 is no longer the premium-priced outlier. At $5 input / $25 output, it now sits within 10% of GPT-5.5 ($5 / $30). The old assumption that "Opus is 3× the price" is out of date as of Anthropic's Opus 4.5 reset.
• Gemini 3 Pro wins short coding tasks on price ($0.16 vs GPT-5.5's $0.39) but its >200K tier doubles input price. For 1M-context workloads, the gap to GPT-5.5 narrows to ~20% cheaper, not 60%.
• GPT-5.5 Batch tier ($2.50 / $15) re-creates old GPT-5.4 standard pricing. For non-realtime production, this is the most defensible cost-per-quality slot in the lineup.

Through TokenMix.ai's unified API, the same gpt-5.5 model is exposed alongside Claude Opus 4.7 and Gemini 3 Pro behind one key. Switching for cost-sensitive workloads becomes a one-parameter change rather than a refactor.

How to Access GPT-5.5 API Today (Direct vs TokenMix)

Two routes, different tradeoffs.

Direct via OpenAI

Requires an OpenAI organization with valid payment method. Tier rate limits apply — Tier 1 accounts get ~30K TPM on gpt-5.5, scaling with usage. No regional cap on availability after launch.

from openai import OpenAI
client = OpenAI()
r = client.chat.completions.create(
    model="gpt-5.5",
    messages=[{"role": "user", "content": "Explain Ramsey numbers."}]
)

Via TokenMix.ai (OpenAI-Compatible Endpoint)

Same SDK, one base URL change. TokenMix.ai routes the request to OpenAI upstream with no markup on GPT-5.5 standard tier (Batch/Flex pricing pass-through), plus aggregated rate limits and a single bill across 170+ models.

from openai import OpenAI
client = OpenAI(base_url="https://api.tokenmix.ai/v1", api_key="tkmx-...")
r = client.chat.completions.create(
    model="gpt-5.5",  # Same model ID
    messages=[{"role": "user", "content": "Explain Ramsey numbers."}]
)

Migration Checklist: GPT-5.4 → GPT-5.5

Most production GPT-5.4 callers can switch to GPT-5.5 with a model-ID change. The real questions are cost and prompt portability.

If your code is hardcoded to gpt-5.4, abstract behind a config flag now. Per TokenMix.ai integration patterns, the cleanest fix is putting the model ID in environment variables and rolling forward when eval data confirms a win.

FAQ

When was GPT-5.5 released?

GPT-5.5 launched on April 23, 2026, with simultaneous availability in ChatGPT, Codex, and the API. GPT-5.5 Instant — the free-tier variant for ChatGPT — followed on May 5, 2026. The internal codename "Spud" was confirmed in OpenAI's launch post.

How much does the GPT-5.5 API cost?

Standard pricing is $5 per million input tokens and $30 per million output tokens. GPT-5.5 Pro costs $30 input / $180 output per million tokens. Batch and Flex tiers cut standard pricing by 50%. Priority tier costs 2.5× standard for guaranteed throughput.

Is GPT-5.5 better than Claude Opus 4.7?

GPT-5.5 wins decisively on Terminal-Bench 2.0 (82.7% vs 69.4%) and dominates long-context retrieval beyond 256K tokens. Claude Opus 4.7 still leads on SWE-Bench Pro Public (64.3% vs 58.6%) and GPQA Diamond (94.2% vs 93.6%). For agentic coding workflows, GPT-5.5 is now ahead. For pure code-completion benchmarks, Opus 4.7 retains the edge.

Is GPT-5.5 better than Gemini 3 Pro?

GPT-5.5 leads on every category except ARC-AGI-1 (Gemini 3 Pro: 98.0% vs GPT-5.5: 95.0%) and BrowseComp (Gemini 85.9% vs GPT-5.5 84.4%). On math (FrontierMath Tier 4: 35.4% vs 16.7%) and tool use (Toolathlon: 55.6% vs 48.8%), GPT-5.5 is clearly ahead. Cost-wise, Gemini 3 Pro at $2/$12 per MTok is 60% cheaper.

What is GPT-5.5's context window?

1M tokens via the API for both gpt-5.5 and gpt-5.5-pro. Codex exposes a 400K-token window. At 512K-1M context, GPT-5.5 holds 74.0% on MRCR v2 8-needle retrieval — up from 36.6% on GPT-5.4. This is the first OpenAI model where 1M-token RAG is practically usable.

Should I migrate from GPT-5.4 to GPT-5.5?

For agentic coding, long-context RAG, or complex tool-use workflows: yes, run an eval. For high-volume short prompts where GPT-5.4 hits accuracy targets: stay on 5.4 unless your benchmarks show a meaningful lift — GPT-5.5 is 2× the per-token cost. Batch and Flex tiers narrow the gap.

Can I use GPT-5.5 through TokenMix.ai?

Yes. gpt-5.5 and gpt-5.5-pro are available through the TokenMix.ai unified API with the same OpenAI-compatible SDK — only the base_url changes. Through the TokenMix.ai platform, the same key calls GPT-5.5, Claude Opus 4.7, Gemini 3 Pro, DeepSeek V4 and 170+ other models with aggregated rate limits.

Is "GPT-6" the same as "GPT-5.5"?

OpenAI confirmed the name is GPT-5.5, not GPT-6. The Spud codename produced the GPT-5.5 release. A future generational version may be named GPT-6, but that has not been announced.

Related Articles

• GPT-5.5 vs Claude Opus 4.7: Real Benchmark Comparison (2026)
• OpenAI API Pricing 2026: Every Model — $0.03/M to $180/M
• Claude Opus 4.7 Review: When to Use It Over GPT-5.5
• Gemini 3 Pro API Pricing & Benchmarks
• How to Switch AI Providers Without Refactoring