Gemini Flash models launched by Google include Gemini 3.6 Flash as the main release, with Gemini 3.5 Flash-Lite and 3.5 Flash Cyber. The Flash series is optimized for speed and for use with AI agents; Gemini 3.6 Flash uses 17% fewer output tokens than 3.5 Flash and prices at $1.50 per million input tokens and $7.50 per million output tokens, down from $9 for 3.5 Flash outputs. The announcement was presented in the context of May 2026 Google I/O, and Google also teased Gemini 4 as a future release.
The launch was accompanied by market-facing analysis that noted pricing and performance changes as well as competitive benchmark comparisons. Alphabet’s stock reaction and other test results were part of that coverage.
Google launched three new AI models: Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber. Gemini 3.5 Pro was held back because it fell short of internal targets, particularly on coding tasks, per Bloomberg. A late-June attempt to fix Gemini 3.5 Pro by updating training data produced disappointing results. Alphabet shares fell roughly 4.4% on the Bloomberg report, erasing about $200 billion in market capitalization in a single session. The last Pro-tier model shipped was Gemini 3.1 Pro in February.
On DeepSWE v1.1, Gemini 3.6 Flash scored 49% compared with 37% for Gemini 3.5 Flash. On MLE-Bench, Gemini 3.6 Flash scored 63.9% versus 49.7% for 3.5 Flash. On OSWorld-Verified, Gemini 3.6 Flash scored 83.0%, compared with Claude Sonnet 5 at 81.2% and GPT-5.6 Luna at 72.6%.
GPT-5.6 Luna scored 67% on DeepSWE v1.1 and 84.7% on Terminal-Bench 2.1. Claude Sonnet 5 topped GDPval-AA v2 with a score of 1607 versus Gemini 3.6 Flash’s 1421. The article noted that Alphabet’s stock reaction and the test results were part of a market-facing analysis of the new Gemini Flash models.
The delay of the Pro-tier, the unsuccessful late-June update, and the market reaction were included in coverage of the new releases. The same coverage also documented Gemini 3.6 Flash’s benchmark improvements over 3.5 Flash and its competitive placements against other models.
Deepseek examined the initial Gemini 3.6 Flash build, identified eleven distinct bugs, and implemented eight targeted fixes that produced a playable game. The applied corrections remedied errors that had prevented proper function, and Deepseek reported the outcome as a decent, working result. The analysis indicated the model’s core reasoning was correct while detailed implementation errors caused the inaccuracies observed earlier. This debugging and testing work was presented within the same market-facing coverage that included benchmark test results and Alphabet’s stock reaction.
