Introducing Gemini 3.8 Live and 3.8 Live Extended Thinking: Google Splits Its Voice Line in Two

Google's Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking arrived on September 15, 2026 — one announcement, two models, and a genuine fork in the product line. The headline number is a 6.6-point gap on Artificial Analysis's Speech to Speech Index, 82.6 to 76.0. The number underneath it is 38. On the same board's τ-Voice agentic task completion measure, the reasoning variant resolves 68.6% of replica customer-service scenarios and the standard model resolves 30.1%. The 6.6 points are what the index says. The 38 points are what happens when you ask either model to actually get something done.

A day on from launch, the fork has a price attached. Both models carry published per-minute audio rates through the Live API, and Artificial Analysis's leaderboard puts cost-per-hour of input audio at $0.84 for the standard model against $3.50 for the reasoning variant — a little over four times as much per hour for those 6.6 index points. That trade is the whole decision, and reading it off the index alone gets it wrong.

What makes this release different from the usual voice-model refresh is that the split is not a size tier. It is a disagreement about what a voice agent's job is. One of these models is a conversational interface. The other is a reasoning system that happens to talk.

Two models, two jobs

Google's own framing is unusually clean about this. Gemini 3.8 Live is described as "built for scale and cost efficiency, combining conversational intelligence with fluid dialogue and visual grounding." Gemini 3.8 Live Extended Thinking is "built for high-complexity tasks, with increased intelligence and multi-step reasoning."

The practical difference shows up in the capability list rather than the spec sheet:

• Gemini 3.8 Live — near real-time visual input processing, automatic mid-conversation switching across 97 supported languages, and background tool/API execution so the model acknowledges a request and keeps talking while the call finishes.

• Gemini 3.8 Live Extended Thinking — reasons and speaks at the same time, narrating progress with cues like "Let me check that…" while a multi-step task runs.

• Shared — all generated audio carries Google's SynthID watermark, both are covered by a model card published under the name gemini-3-8-audio, and both now have documented API model IDs: gemini-3.8-live and gemini-3.8-live-extended-thinking.

The Extended Thinking model is not simply 3.8 Live with a longer thinking budget. It is the one that can hold a conversation and a task plan simultaneously without the conversation stalling — which is precisely the thing that makes voice agents fail in production. A user does not want silence while the agent looks something up. They want the agent to keep talking.

The board says 6.6. The components say 38

The Speech to Speech Index is a weighted average of four underlying results: Speech Reasoning, measured by Artificial Analysis's own Big Bench Audio set; Agentic Performance, measured by running the τ-Voice customer-service benchmark; Arena Preference, taken from the Speech Agent Arena; and Task Success Rate. Only models with all four components receive an index score, which is why some rows on that board are blank.

Pulled apart, the two new models do not look like a 6.6-point pair at all:

• Speech to Speech Index — Gemini 3.8 Live Extended Thinking (High) 82.6 vs Gemini 3.8 Live 76.0

• Agentic performance (τ-Voice task completion) — 68.6% vs 30.1%

• Speech reasoning (Big Bench Audio) — 98% vs 92%

• Conversational dynamics — 91.9% vs 96.1%

• Arena preference (Elo) — 990 vs 1083

• Task success rate — 89.1% vs 93.2%

• Time to first audio — 1.35s vs 1.18s

• Cost per hour of input audio — $3.50 vs $0.84

Read it honestly and the standard model wins four of eight lines. It is faster to first audio, rated higher by the arena, more likely to complete a task, and rated better on conversational dynamics. It is also a third of the price. What it cannot do is finish the job when the job has steps: 30.1% against 68.6% is the largest single gap anywhere in this release, and it sits on the one axis that separates a voice agent from a voice interface.

Two caveats before that table gets treated as settled. The arena preference figure used in the index is frozen at the point a model becomes eligible for publication, so it does not track the live Elo on the Speech Agent Arena chart — read it as a snapshot, not a running score. And the τ-Voice margins at the top of the board are thin enough to be noise: the reasoning variant's 68.6% leads GPT-Live-1 at 67.9% (Astra backend, medium effort) by 0.7 points, with GPT-Live-1 (Sol, low) at 59.3% and Grok Voice Think Fast 2.0 High at 56.5% behind it. The 38-point gap between the two Gemini models is not thin. The 0.7-point lead over GPT-Live-1 is.

Where it sits on the index

In the capture below, taken on September 16, 2026, the top of the board reads like this:

• Gemini 3.8 Live Extended Thinking (High) — 82.6 (first place)

• GPT-Live-1 (Astra backend, medium effort) — 81.5

• Grok Voice Think Fast 2.0 High — 81.3

• GPT-Live-1 (Sol backend, low effort) — 80.1

• Gemini 3.8 Live — 76.0

• GPT-Realtime-2.1 High — 73.9

• Gemini 3.1 Flash Live High — 71.5

That is the case for the split in one screen. The Extended Thinking variant takes the top spot, running at the board's (High) reasoning-effort label — the same convention it uses for Grok Voice Think Fast 2.0 High and GPT-Realtime-2.1 High, and the reason the "(High)" suffix you may see attached to this model in coverage is a configuration label rather than a separate release. The standard variant lands fifth — below two GPT-Live-1 configurations and below Grok. Google's own blog describes the standard model as "highly cost-effective" and notes it "secured a second place in the Speech Agent Arena," which is a different board measuring a different thing. Both statements are true. Read together they say: 3.8 Live is a very good conversational model at a very good price, and it is not the frontier of voice intelligence.

The leak that got there first

The models were spotted on a Google Cloud quota page the day before the announcement, and we covered that sighting at the time as an unconfirmed slug with no model card, no pricing, and no Google confirmation. That post was right about the status and wrong about the timeline — the confirmation arrived within roughly 24 hours.

The lesson is worth recording, because it cuts against the usual instinct. A leaked slug one day before launch is a launch. A leaked slug with no corroborating paperwork for eight weeks is a different thing entirely. The signal that separated them was not the slug; it was the presence of a quota page, which only exists when capacity has been provisioned.

What it costs, per minute

At announcement, Google's materials did not separate the two models on price — the standard model was given a rate and the reasoning variant was not, which is what the scoreboard above records. That gap has since closed. Both models now carry published pricing through the Live API: $0.005 per minute of audio input and $0.018 per minute of audio output, which Google confirmed on September 16 as the two rolled out. The same published rate card covers the Gemini 3 Live tier's token pricing — $0.75 per 1M text tokens in and $4.50 out, $3.00 per 1M audio tokens in and $12.00 out, and $1.00 per 1M for image or video input — though those token rows are published for the tier rather than per model, so read them as the tier's rate card and not as a quote for a specific variant.

Artificial Analysis has also filled in the number that matters most for planning. Its leaderboard now carries a cost-per-hour-of-input-audio column — the cost to complete a fixed 40-question Big Bench Audio subset, normalised to an hourly rate — and both new models have a figure there:

• Gemini 3.8 Live — $0.84/hour of input audio, the lowest paid rate on that board

• Gemini 3.8 Live Extended Thinking (High) — $3.50/hour, still below both models it beats on quality

• Grok Voice Think Fast 2.0 High — $4.80/hour

• GPT-Live-1 (Astra backend, medium effort) — $5.83/hour

• GPT-Realtime-2 (High) — $4.14/hour

• GPT-Realtime-2.1 High — $10.75/hour

One note on the captures above, since the costs panel in the leaderboard screenshot predates these two rows: neither $0.84 nor $3.50 appears in it, and its cheapest bar is $1.42. The figures in the list are read from the board's summary table as it stands on September 16, 2026, not from that image. The index values shown in the image are unaffected and match the table.

Read the two new numbers against the components above and the release stops being a two-model announcement and becomes a single decision with a published exchange rate. The reasoning variant buys 6.6 index points, six points of speech reasoning and 38.5 points of agentic task completion for a little over four times the hourly audio cost. Whether that is worth paying depends entirely on what your agent is doing — and the shape of the answer changed once the components were published. A voice agent whose job is to finish a multi-step task is buying the single largest improvement in this release, at $3.50 an hour while undercutting GPT-Live-1 Astra by about 40% and Grok Voice Think Fast 2.0 High by roughly a quarter, and scoring above both. A voice agent whose job is to converse — answer, hold a thread, take a message, be pleasant — is buying very little with extra reasoning depth, and at $0.84 an hour it is buying the cheapest competent voice model anyone currently publishes.

That is the shape of the list. Google has priced its reasoning model below the two models it beats, and priced its standard model below everything on the board. If you are running voice minutes at volume, this is the number that moves your bill, not the index score — and the fact that the two variants sit 4× apart means picking which one to call is the single largest cost lever in the stack.

"Private preview" is doing real work in that sentence

Neither model is generally available in the contractual sense, but the developer surface is open and now priced, which changes what you can plan on:

• Gemini 3.8 Live — developers get it in the Gemini API, the Live API and Google AI Studio under the documented ID gemini-3.8-live; enterprises get private preview in Gemini Enterprise, with Gemini Enterprise for Customer Experience "coming soon"; everyone gets it in Search Live.

• Gemini 3.8 Live Extended Thinking — the same developer surface under gemini-3.8-live-extended-thinking, plus a wider consumer path: Gemini Live, Docs Live for Google AI Pro and Ultra subscribers, and Gmail Live and Keep Live for all Google AI subscribers.

What is still missing is the part an enterprise needs: a general-availability commitment, a published uptime figure, and a rate Google has promised to hold. If you need those, wait. If you are prototyping, the path is open today with a real rate card behind it, which is a materially better position than a preview with no published price. Building the integration now is reasonable; putting a revenue-critical phone line on an availability target Google has not committed to is not, and the fix for that is described below.

Where the vendor claims end

Google published per-component figures of its own alongside the launch — 68.6% on τ-Voice, 35.1% on Sierra's τ³-Banking leaderboard, and 97.7% on Big Bench Audio, all credited to the Extended Thinking model. Those now divide into two groups, and the split matters.

Two of the three line up with measurements Artificial Analysis runs itself under its own harness. Its τ-Voice agentic component reads 68.6% for this model, against 67.9% for GPT-Live-1 Astra and 56.5% for Grok Voice Think Fast 2.0; its Big Bench Audio speech-reasoning component reads 98%, against Google's 97.7%. Artificial Analysis describes τ-Voice and Big Bench Audio as its own benchmarks, run across three trials where available. So the τ-Voice and reasoning numbers now sit on a public board you can go and read, not only in Google's announcement — treat them as corroborated rather than settled, since the exact figures Google quoted may well be that same run rather than a second one.

The Sierra number has no such backing and remains a vendor claim: 35.1% on Sierra's τ³-Banking leaderboard against 32.0% for GPT-Live-1 Astra and 16.5% for xAI-Realtime. It is also the number worth sitting with, because 35.1% means the leading voice model on this board fails roughly two of every three realistic banking task-completion attempts. Google also cites a ServiceNow EVA-Bench run performed on the Live API in Gemini Enterprise Agent Platform, and describes both models as pushing the Pareto frontier for complex workflows — neither of which has an independent reading attached.

There is a precedent worth holding onto here. xAI reported a Speech to Speech Index figure of 82.9 for Grok Voice Think Fast 2.0 in July. On the board captured above, that model sits at 81.3. Vendor-reported index scores do not always survive contact with a live leaderboard — usually because the index is revised, sometimes because the configuration tested was not the one that shipped. Verify the τ-Voice and Big Bench numbers on your own traffic, and read the agentic column as the one to test hardest.

The layer these agents actually run on

Both models sit in front of a backend. Background tool execution, multi-step task planning, and the delegation pattern that every serious voice agent uses all resolve into ordinary text-model calls — and that is the layer with the most cost variance and the least lock-in.

OrcaRouter carries 190 models from a single key at provider list price with no markup, which means a price cut from an upstream lab is live on our side the same day rather than at the next contract renewal. For a voice stack where the front-end is a preview model, the two useful properties are that the delegation target can be swapped without touching the voice integration, and that automatic failover keeps a call alive when a backend errors or times out. To be precise about what we do and do not host: the Gemini 3.8 Live endpoints are not on our router — those come from Google's own API — but the text models these agents hand work off to very often are, Gemini 3.8 Flash among them.

What to watch next

Three things will settle the open questions in this release. The first is general availability and a rate Google has committed to hold — the prices are published now, but preview pricing has a habit of moving, and a rate card is not a contract. The second is whether the arena and task-success columns hold up, because the standard model's 1083 Elo and 93.2% task success are the strongest argument against paying 4× for the reasoning variant, and the index's arena figure is frozen at eligibility rather than live. The third is independent runs on the agentic gap itself: 68.6% against 30.1% is the largest claim in the release and the one most worth reproducing, because everything else about the two models is close.

Until then, the honest summary runs on two numbers rather than one. Gemini 3.8 Live Extended Thinking is the best-scoring voice model on the public board, leads the agentic component outright, and undercuts the two models directly behind it on hourly cost. Gemini 3.8 Live is the cheapest competent voice model on that board, is preferred by the arena and more reliable on shallow tasks, and sits 6.6 points back overall with a hard ceiling at the one thing agents get hired for. Google has shipped the same fork twice, four times apart on price, and made the choice unusually easy to price — provided you read the components and not just the index.