Google and DeepSeek each released new models on August 13, both aimed at coding and multi-step agent workflows. Google announced Gemini 3.7 Flash, and DeepSeek moved its V4 Pro model out of preview into general release.
Gemini 3.7 Flash
Gemini 3.7 Flash arrives roughly three weeks after Gemini 3.6 Flash. Google attributes the short interval to developer feedback and internal algorithmic work.
Google’s published benchmarks, reported by VentureBeat, show 43.6% on FrontierCode 1.1 Main against 34.4% for the previous model, 65.3% on DeepSWE v1.1 against 49.0%, and 30.4% on AutomationBench against 17.0%. These figures are vendor-supplied, and several of the evaluations originate with commercial AI companies rather than independent bodies.
Google has also changed how it describes the model’s behaviour. Documentation for 3.6 Flash emphasised reducing reasoning steps and tool calls. For 3.7, Google describes the goal as more deliberate planning followed by cleaner execution.
Introductory pricing is $0.75 per million input tokens and $3.75 per million output tokens through the end of the year, half the rate for 3.6 Flash. The model is also available in Gemini Spark, Google’s subscription agent service for AI Pro and Ultra subscribers in more than 160 countries.
Gemini 3.5 Pro remains unreleased
Google’s flagship model, announced at I/O on May 19 and originally scheduled for June, has still not shipped. The Flash line has now reached version 3.7 while the Pro line remains at 3.5 and unreleased, with Gemini 3.1 Pro still the current flagship.
DeepSeek V4 Pro
DeepSeek released V4-Pro-0813 on August 13, ending a preview period that began in April. The company identifies agent capability as the primary improvement. The model is available through DeepSeek’s API, app, and web interface, according to Reuters.
DeepSeek’s model documentation describes a mixture-of-experts architecture with approximately 1.6 trillion total parameters, about 49 billion active, and a one-million-token context window.
During the preview period, DeepSeek’s smaller V4 Flash model outperformed the V4 Pro preview on several independent tests.
DeepSeek raises API prices
DeepSeek announced increases of 50% to approximately 1,100% across its V4 models, depending on model, token type, and time of day, as reported by Reuters. DeepSeek’s pricing page sets the change at 16:00 UTC on August 16, which is midnight on August 17 in Beijing; both dates appear in coverage.
The new structure introduces peak and off-peak rates. Peak windows run 01:00 to 04:00 and 06:00 to 10:00 UTC, with off-peak billed at half the peak rate. Quartz reports V4-Flash output tokens moving to $1.32 per million at peak and $0.66 off-peak, from a flat $0.28.
Effect on developers
Google’s introductory rate expires December 31, after which pricing returns to the previous Flash tier. DeepSeek’s peak and off-peak split makes scheduling relevant to cost: batch processing, evaluation runs, and backfills can be moved into off-peak windows.
Agent workloads consume tokens across many planning steps and tool calls rather than once per user turn, which increases the effect of any rate change on total spend.
Leave a Reply