GPT-5 reportedly dropping next week with 1M token context window and dynamic reasoning
Jul 29, 2025
Key Points
- OpenAI is reportedly releasing GPT-5 next week with a 1M token context window, dynamic reasoning, and MCP support, though Polymarket prices the July 31st launch at just 7% probability.
- OpenAI has redefined GPT-5 away from raw pretraining scale, instead bundling its strongest base model, best tools, and most usable interface into a single offering.
- Dylan Patel said an earlier Orion candidate model "made him laugh" with "good model smell," but o3's reasoning capabilities proved more valuable to users than raw scale.
Summary
OpenAI is reportedly releasing GPT-5 next week, according to a report from DD Doss citing an anonymous source with what he describes as reasonable proof. The claimed specs include a 1M token context window, 100,000 output tokens, MCP support, parallel tool calls, and dynamic short and long reasoning that integrates code interpreter and other tools. Internal code names reportedly include 03 alpha, nectarine, lobster, lobster mini, starfish, and nano.
Polymarket pricing reflects skepticism on the immediate timeline. GPT-5 released by July 31st is priced at 7%, but August 15th sits at 74%. By December 31st, the probability reaches 97%, suggesting market consensus expects the release within the next several months rather than imminently.
GPT-5 no longer necessarily means a massive leap in pretraining scale. OpenAI previously ran a candidate model called Orion that was supposed to be GPT-5, a massive training run that tested well internally. Dylan Patel reportedly said it "made him laugh" and had "good model smell," but it proved less useful than the o3 and o3 Pro reasoning models, which excel at tool use and structured reasoning.
OpenAI appears to be interpreting the version bump differently. Rather than tying GPT-5 to a specific training run, the release appears to be a consolidation of OpenAI's best capabilities: strongest base model, best tools, best systems, best interface bundled into a single offering. GPT-4.5, which was rebranded from an earlier GPT-5 candidate, was pushed behind a non-default UI, ran slower, but users found it "funnier" and felt it had "more personality." The pattern suggests OpenAI is optimizing for usability and consistency rather than raw pretraining size.
Expectations are extremely high for any GPT-5 release. What users actually want is something usable.