GPT-5.2 fit teams needing agentic tool-calling reliability across complex, multi-turn workflows more than raw leaderboard rank, where it placed third on LMArena's human-preference leaderboard at launch. It's deprecated now, so any new integration should target GPT-5.5, OpenAI's designated successor, instead of building on a retired endpoint.
GPT-5.2 is OpenAI's fourth GPT-5 variant, the first to triple the family's context window on a Mixture-of-Experts architecture that routes each token through specialized expert networks. It reached 88.7% on CharXiv chart comprehension with Python tool access, before OpenAI deprecated it in June 2026.
Where it sits
- $4.81/M$ per 1M tokensBlended price (3:1)Lower is better#45 / 65peer median $1.71/Mvendor price, checked by HokAI
- 71 tok/stokens/sOutput speedHigher is better#25 / 40peer median 87 tok/scited: Artificial Analysis
- 80%% solvedSWE-bench VerifiedHigher is better#12 / 29peer median 78.5%per source, see benchmark scores
- 92.4%% correctGPQA DiamondHigher is better#13 / 45peer median 88.9%per source, see benchmark scores
Pricier than 69% of the 65 GA models with a published price, mid-pack on SWE-bench Verified (rank 12 of 29), and one of 23 that document a zero-data-retention option. Ranked against GA models; this record is not GA.
Ranks are against GA models on HokAI that publish the same figure; ties share a rank.
Provider: OpenAI · Family: GPT-5
Context window: 400,000 tokens · Max output: 128,000
Input modalities: text, image, audio, video, pdf, tool-calls · Output: text, tool-calls
About GPT-5.2
GPT-5.2 is OpenAI's fourth release in the GPT-5 product family, launched on December 11, 2025. It positions above GPT-5.1 in the lineup and was eventually succeeded by GPT-5.3 and GPT-5.5. The model uses a Mixture-of-Experts (MoE) transformer architecture, routing each token to a specialized subset of expert networks rather than activating all parameters per inference. This reduces compute cost per query while maintaining large total model capacity. OpenAI has not published the parameter count, but inference cost patterns and MoE activation ratios place it in the 2-to-5 trillion total parameter range. The model ships in three service tiers: GPT-5.2 Instant (low latency, no extended reasoning), GPT-5.2 Thinking (standard and extended chain-of-thought before responding), and GPT-5.2 Pro (maximum quality with an unlimited thinking budget). A coding-specialized companion, GPT-5.2-Codex, launched the same day.
On release, GPT-5.2 posted the strongest benchmark results in the GPT-5 family to that point. It achieved 80.0% on SWE-bench Verified (agentic coding), 92.4% on GPQA Diamond (graduate-level science reasoning, a 4.3-point gain over GPT-5.1's 88.1%), and a perfect 100% on AIME 2025 (competition mathematics) using the Thinking tier without external tools. On FrontierMath, a contamination-resistant advanced math benchmark, GPT-5.2 scored 40.3%, ahead of Claude Opus 4.5 at 37.6% and Gemini 3 Pro at 31.1%. Visual reasoning on chart interpretation roughly halved error rates vs GPT-5.1. On the LMArena human-preference leaderboard, the model reached an Elo rating of 1402 as of May 2026, ranking third behind Claude Opus 4.6 at 1418 and Gemini 3.1 Pro at 1406. Confidence intervals across the top tier overlapped, meaning cost and latency should drive selection more than raw leaderboard position.
The 400,000-token context window is the headline architectural change over GPT-5.1, which topped out at 128,000 tokens. The max output cap is 128,000 tokens. OpenAI reported over 99% recall accuracy on internal needle-in-haystack tests at full 400K depth for the Thinking tier. Independent evaluations found some degradation above 350,000 tokens, with recall for instructions placed at the very start of long prompts dropping from 99% to around 88% near the 400K ceiling. For comparison, Claude Opus 4.5 offered 200K context and Gemini 3 Pro offered a 1M-token window at reduced pricing; GPT-5.2 sat between those two on size but demonstrated stronger recall quality in the 100K-to-300K range in independent tests. For full-codebase review, multi-document contract analysis, and long research-paper synthesis without chunking, the 400K window covered the majority of real enterprise workloads.
GPT-5.2 handles text, image, audio, and video inputs within a single API request. Video understanding scored 90.5% on Video-MMMU vs Gemini 3 Pro's 87.6%. Chart comprehension on CharXiv with Python reached 88.7%. OpenAI reported 98.7% accuracy on the Tau2-bench Telecom evaluation, an agentic test covering complex, long-horizon, multi-step tasks, and cited this as evidence the model executes cleanly off a simple one-line system prompt where previous models required elaborate scaffolding. The model generates structured JSON natively and supports parallel tool calls in a single completion. Computer use capabilities and web search are accessible through the ChatGPT product interface, but are not exposed as raw API endpoints in GPT-5.2.
GPT-5.2's API pricing sat well above GPT-5's sub-$1-per-million-token launch rate, reflecting a quality-tier premium for the added context length and reasoning depth. Output tokens cost several times more than input tokens, a ratio developers frequently flagged as expensive for generation-heavy workflows such as bulk code generation. Prompt caching cut the effective cost of repeated long-context calls once a request passed the minimum cache-match threshold. Exact rates, cached-input pricing, and representative cost examples appear in the pricing table and cost FAQ below.
Access during the active period was through the OpenAI direct API at platform.openai.com and through Microsoft Azure OpenAI Service under the commercial arrangement in place through 2025. GPT-5.2 did not receive a standalone AWS Bedrock listing before its deprecation; Bedrock support for the GPT family launched later with GPT-5.4 and GPT-5.5. No confirmed availability on Google Vertex AI, Together AI, or Fireworks AI is documented. The model supported OpenAI's Python, TypeScript, and other official SDKs using the same chat completions endpoint structure as earlier GPT-4 models, meaning migrations from GPT-4 and GPT-5.1 required minimal code changes.
OpenAI trained GPT-5.2 using reinforcement learning with extended chain-of-thought, integrating the reasoning techniques from the o1 and o3 families into the main GPT-5 product line. The knowledge and training data cutoff is August 31, 2025. The system card, published December 11, 2025 as an addendum to the GPT-5 system card, documents safety training covering agentic task refusal, resistance to prompt injection in multi-tool environments, and deployment at a GPT-4-class uplift tier under OpenAI's Preparedness Framework. The companion GPT-5.2-Codex system card added agent sandboxing and configurable network access as product-level mitigations for coding agent deployments. OpenAI does not train on API inputs by default; enterprise accounts can enable a zero-data retention option via agreement.
GPT-5.2 was deprecated from the OpenAI API on May 8, 2026, with developer notifications issued simultaneously. It was retired from ChatGPT on June 12, 2026, with active conversations automatically migrating to the corresponding GPT-5.5 tier. GitHub Copilot deprecated GPT-5.2 and GPT-5.2-Codex on June 5, 2026, retaining GPT-5.2 only in Copilot code review. The successor, GPT-5.5, improved SWE-bench Verified and reduced output token pricing, addressing the two main developer complaints about GPT-5.2. Teams still on GPT-5.2 integrations should migrate to GPT-5.5 or GPT-5.4 to avoid service disruptions.
Pricing
$1.75 per 1M input tokens, $14.00 per 1M output tokens. Prompt caching at $0.18 per 1M cached input (90% discount, minimum 1,024-token match). Model is deprecated; migrate to GPT-5.4 or GPT-5.5 for continued access.
What a real job costs
| Job | Input | Output | Total |
|---|---|---|---|
| Summarise a 20-page PDF | $0.052 | $0.014 | $0.067 |
| Support reply | $0.0035 | $0.0042 | $0.0077 |
| One coding agent run | $0.350 | $0.280 | $0.630 |
Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.
Key Features
- Largest Context Window in the GPT-5 Family: The first GPT-5 variant to move past the 128,000-token ceiling shared by GPT-5 and GPT-5.1, enabling full-codebase reviews and multi-document legal analysis in one API call.
- Three-Tier Reasoning Design: Ships as Instant (low latency, no chain-of-thought), Thinking (extended reasoning before response), and Pro (unlimited thinking budget), so cost and latency can be matched to task complexity per request.
- Full Multimodal Input: Accepts text, images, audio, and video in a single request; scored 90.5% on Video-MMMU and roughly halved error rates vs GPT-5.1 on chart and software UI comprehension.
- 98.7% Tool Calling Accuracy: Scores 98.7% on Tau2-bench Telecom, a multi-turn agentic benchmark, meaning complex agent loops with many sequential tool calls work with a simple one-line system prompt.
- GPT-5.2-Codex Companion: A coding-specialized variant released the same day, with agent sandboxing, configurable network access, and tighter code-execution safety controls for autonomous software development pipelines.
Pros
- Posted the strongest math and science reasoning scores in the GPT-5 family to that point, using its Thinking tier without external tools.
- Context window large enough to cover full codebases and multi-document legal reviews in a single pass, with strong internal recall at typical working depths.
- Tool-calling accuracy was strong enough on multi-turn agentic benchmarks to work reliably off a simple system prompt, without elaborate scaffolding.
Cons
- Deprecated June 2026 after only 6 months; new projects should target GPT-5.4 or GPT-5.5 to avoid API cutoff.
- Output pricing ran several times higher than input pricing, making generation-heavy workloads like bulk code generation comparatively expensive.
- No native audio output and no direct computer-use API endpoint; voice pipelines require a separate TTS call.
Benchmarks
- AIME 2025: 100% vendor-reported · 11 Dec 2025 — Competition-level maths problems from the 2025 exam, % solved.
- HumanEval: 97.4% vendor-reported · 11 Dec 2025 — Small programs that must pass hidden tests, % passing.
- LMArena Elo: 1402 vendor-reported · 11 Dec 2025 — Rating from blind human votes on which answer is better.
- GPQA Diamond: 92.4% vendor-reported · 11 Dec 2025 — PhD-level science questions that are hard to search for, % correct.
- LMArena rank: #3 vendor-reported · 11 Dec 2025 — Position on the blind human-preference leaderboard; #1 is best.
- SWE-bench Verified: 80% vendor-reported · 11 Dec 2025 — Real GitHub issues fixed end to end, % solved.
- AA Intelligence Index: 51 cited: Artificial Analysis · 18 Jun 2026 — Composite of 10 evaluations run by Artificial Analysis, 0 to 100.
- AA blended price: $1.87/M cited: Artificial Analysis · 18 Jun 2026 — Price per 1M tokens at a 3:1 input to output blend, as listed by Artificial Analysis.
- Output speed: 71 tok/s cited: Artificial Analysis · 18 Jun 2026 — Median tokens written per second as measured by Artificial Analysis.
A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.
Frequently Asked Questions
How much does GPT-5.2 cost in 2026?
GPT-5.2 was priced at $1.75 per 1M input tokens and $14.00 per 1M output tokens through the OpenAI API, with cached input at $0.18 per 1M after a 1,024-token minimum cache match. Artificial Analysis calculated a blended rate of $1.87 per 1M tokens using a 7:2:1 weighting. The model is deprecated as of 2026 and no longer billable through OpenAI's standard signup flow; teams should budget for GPT-5.5's current API pricing instead.
Is GPT-5.2 open source or proprietary?
GPT-5.2 is fully proprietary and closed-weight: there is no self-hosting path, and all access ran through OpenAI's direct API or Azure OpenAI Service during its active window. It carried SOC 2 Type II certification and HIPAA-eligible configurations via a Business Associate Agreement, with GDPR compliance available under OpenAI's Data Processing Agreement. OpenAI's separate open-weight releases, gpt-oss-20b and gpt-oss-120b, share no code or weights with GPT-5.2.
What are GPT-5.2's closest competitors?
Since GPT-5.2 is deprecated, the practical alternatives for new work are GPT-5.5, its official successor with materially better coding benchmarks and lower output pricing, and GPT-5.4, which added a native Computer Use API GPT-5.2 never exposed. Outside OpenAI, Claude Opus 4.6 remains a directly comparable frontier option that is still receiving API traffic.
What separates GPT-5.2 from Claude Opus 4.6?
Claude Opus 4.6 scores 80.8% on SWE-bench Verified and offers agent teams on a much larger context window than GPT-5.2 ever had. GPT-5.2 matched Opus 4.6 closely on general coding ability at launch, but OpenAI retired the GPT-5.2 endpoint in June 2026 while Opus 4.6 continues to receive API traffic.
How do you migrate from GPT-5.2 to GPT-5.5?
GPT-5.5 uses the same chat completions endpoint structure as GPT-5.2, so most integrations only need a model-name swap and a re-test of prompts. Confirm behavior at your typical prompt length, since GPT-5.5's expanded context window has different long-context recall characteristics than GPT-5.2's. Expect meaningfully higher token pricing than GPT-5.2's original rate; see the pricing FAQ above for exact figures.
Top Alternatives
- GPT-5.5: Pick GPT-5.5 for OpenAI's current flagship: a 1,000,000-token context window and pricing well below GPT-5.2's retired rate.
- GPT-5.4: Pick GPT-5.4 for a native Computer Use API GPT-5.2 never shipped, at a lower per-token cost than GPT-5.2's original price.
- Claude Opus 4.6: Pick Claude Opus 4.6 for continued availability and an 80.8% SWE-bench Verified score, while GPT-5.2's API went dark in June 2026.