OpenAI Launches GPT-5.2: First Model to Hit 100% on AIME Math

Image Credit: Jacky Lee

OpenAI said on Dec 11 it began rolling out GPT 5.2 across ChatGPT and its developer platform, introducing new model variants and updated pricing as it faces intensifying competition from rival AI systems marketed for coding, research, and office style productivity work.

Reuters reported the launch followed an internal “code red” push ordered by chief executive Sam Altman earlier in December, prompted by competitive pressure after Google rolled out Gemini 3 in November.

Rollout in ChatGPT and Model Options

OpenAI said it is rolling out GPT 5.2 Instant, GPT 5.2 Thinking, and GPT 5.2 Pro in ChatGPT, starting with paid plans (Plus, Pro, Go, Business, Enterprise) and deploying gradually to manage performance and reliability. OpenAI added that GPT 5.1 will remain available to paid users inside ChatGPT for three months under legacy models before it is removed from ChatGPT.

OpenAI’s ChatGPT documentation also describes GPT 5.2 Auto as a routing option that can choose between Instant and Thinking depending on the request, while release notes describe changes to how free and Go users access reasoning, aimed at giving users more direct control over when to use deeper reasoning.

Developer Access, Model Specs, and New Controls

For developers, OpenAI lists GPT 5.2 as its flagship model for coding and agentic tasks, with a 400,000 token context window, up to 128,000 output tokens, and a 31 Aug 2025 knowledge cutoff. OpenAI also lists GPT 5.2 Pro as a Responses API only option designed for harder requests that may take longer to complete.

OpenAI’s API guide for the GPT 5 family says GPT 5.2 adds an xhigh reasoning effort level, concise reasoning summaries, and “compaction” features intended to manage context more efficiently in longer workflows.

Pricing and Legacy Availability

OpenAI’s release note lists GPT 5.2 at USD 1.75 per 1 million input tokens and USD 14 per 1 million output tokens, with discounted cached inputs, and it shows higher pricing for the Pro tier model in the API. OpenAI said it has no current plans to deprecate GPT 5.1, GPT 5, or GPT 4.1 in the API and would provide advance notice for any future changes.

Reuters separately reported OpenAI will keep earlier models available to API customers while GPT 5.2 rolls out, a strategy that can help developers avoid forced migrations while they validate behaviour, cost, and reliability.

The 100 Percent Benchmark Result

In its benchmark table, OpenAI reported GPT 5.2 Thinking scored 100.0% on AIME 2025 with no tools, a perfect score on that competition math evaluation under OpenAI’s stated setup. In the same table, OpenAI lists GPT 5.1 Thinking at 94.0% on the same test, meaning this is the first time OpenAI has published a perfect AIME 2025 no tools result for its GPT 5 line in its own release benchmarks.

However, cross vendor comparisons depend on test conditions. Google DeepMind’s published Gemini 3 Pro benchmark table lists 95.0% on AIME 2025 with no tools, and 100.0% with code execution, illustrating how tool access can change reported results.

Safety and Reliability Disclosures

OpenAI published a system card update for GPT 5.2 describing baseline safety evaluations across areas including disallowed content, jailbreak robustness, prompt injection, and hallucinations, alongside notes on limitations of evaluation coverage and the need to anticipate new attack patterns.

Competitive Context

Google’s Gemini 3 launch in November positioned the model as widely available across Google products and developer channels, underscoring the broader market shift toward AI systems that can plan tasks, use tools, and produce work outputs rather than only generate text.

Reuters also reported that alongside the GPT 5.2 launch, Disney announced a USD 1 billion investment in OpenAI and a deal allowing OpenAI to use certain Disney characters in its Sora video generator, linking the model release to wider commercial competition over media, creative tooling, and licensing.

License This Article

Source: Reuters, OpenAI, OpenAI Platform, OpenAI CDN, OpenAI Help

$5.00

$10.00

$20.00

$30.00

Custom Amount

Please enter an amount

$

One-Time Donation Monthly Donation Quarterly Donation Annual Donation

3% Cover the Fee

Support us to keep delivering insightful content on AI development. Your contribution fuels our mission to inspire and inform! Support us to keep delivering insightful content on AI development. Your contribution fuels our mission to inspire and inform!