GEMINI 3.5 PRO STATUS — JULY 21, 2026
● Status: NOT launched — third delay confirmed July 17, 2026
● Reason: Newer checkpoints underperform older ones; hallucinations unresolved on internal evals
● Google's fallback plan: Gemini 3.6 Flash as stopgap; possible skip to Gemini 4.0 Flash
● Claude Sonnet 5: Launched June 30, 2026. Fully live. $2/$10/M intro through August 31. 63.2% SWE-bench Pro. 1M context.
● Practical advice: Do not build around Gemini 3.5 Pro's expected specs. Use Claude Sonnet 5 or GPT-5.6 Terra for mid-tier work now.
Claude Sonnet 5 — Full Specs
| Model |
Input /1M |
Output /1M |
Context |
SWE-bench Pro |
Terminal-Bench |
Status |
| Claude Sonnet 5 |
$2 (intro) |
$10 (intro) |
1M |
63.2% |
78.4% |
Live |
| Gemini 3.5 Pro |
TBC |
TBC |
TBC |
TBC |
TBC |
3rd delay |
The Three Delays — What Went Wrong at Google
Gemini 3.5 Pro has been delayed three times. The first delay pushed the launch from Q2 to early July. The second delay pushed it to July 17. The third delay confirmed July 17 with no new target date. The reason reported for each delay is consistent: newer model checkpoints are underperforming older ones on internal evaluations, and hallucination rates on key tasks remain unresolved. This is a meaningful signal about the difficulty of the capability jump Google is attempting. Gemini 2.5 Pro was a strong model — Gemini 3.5 Pro needs to clear a high bar to justify the version number increment, and it is apparently not clearing it yet.
Google's fallback strategy involves two options: ship Gemini 3.6 Flash as a faster, cheaper stopgap that does not need to beat Gemini 3.5 Pro benchmarks, or skip straight to Gemini 4.0 Flash and retire the 3.5 Pro roadmap entirely. Neither option involves shipping Gemini 3.5 Pro on a timeline anyone can plan around in July 2026.
Why Claude Sonnet 5 Is the Rational Default Right Now
Claude Sonnet 5 at $2/$10/M intro (through August 31) gives you 63.2% SWE-bench Pro accuracy, 1M context, and a model that is live, documented, and available via API today. If you were planning to benchmark Gemini 3.5 Pro against your production workloads before adopting it, that benchmark cannot happen — the model is not available. Claude Sonnet 5 can be benchmarked today. GPT-5.6 Terra ($2.50/$15/M) is the other live alternative in the same tier. Gemini 3.5 Pro remains a future consideration, not a present option.
If Gemini 3.5 Pro Ships — What to Expect
When Gemini 3.5 Pro eventually launches, the comparison to Claude Sonnet 5 will be straightforward: Google's mid-tier flagship vs Anthropic's mid-tier. Based on Gemini's historical pricing, expect Gemini 3.5 Pro at $3-7/M input and $15-35/M output — likely above Sonnet 5's intro price but potentially competitive with the standard $3/$15/M rate Sonnet 5 will reach after August 31. Benchmark performance is genuinely unknown until Google publishes — the delay suggests it may not beat Sonnet 5 on standard coding benchmarks, but could lead on multimodal tasks where Google has historically been stronger.
What to Use Now
Best mid-tier model available right now: Claude Sonnet 5 ($2/$10/M intro). Live, benchmarked, 1M context, 63.2% SWE-bench Pro. Intro pricing ends August 31 — build and test your Sonnet 5 workflows now while pricing is lowest.
Best mid-tier on Terminal-Bench: GPT-5.6 Terra ($2.50/$15/M). 87.1% Terminal-Bench 2.1 vs Sonnet 5's 78.4%. If Google Gemini's ecosystem is important to you, wait for 3.6 Flash — but do not wait for 3.5 Pro, which has no reliable launch timeline.
Watch for Gemini 3.6 Flash instead. Based on Google's own signalling, 3.6 Flash is more likely to ship in the next 2-4 weeks than 3.5 Pro. Flash tier models from Google have consistently delivered strong price-to-performance ratios. That is the Gemini product worth waiting for right now.
Last updated July 21, 2026. This page updates when Gemini 3.5 Pro officially launches. Related: Gemini 3.5 Pro third delay — full story → · Gemini 3.6 Flash — what we know → · Claude Sonnet 5 vs GPT-5.6 Terra →