Claude Fable 5.1 vs Gemini 3.1 Pro: A Nine-Day-Old Flagship vs a Seven-Month Preview

Ask the wrong question about Claude Fable 5.1 and Gemini 3.1 Pro Preview and you will get a boring answer: the new Anthropic flagship is more capable, the Google model is cheaper, case closed. The question that actually decides this matchup is a question of age. Claude Fable 5.1 shipped on September 1, 2026, making it nine days old as this is written. Gemini 3.1 Pro Preview shipped on February 19, 2026, and Google still labels it a preview — which makes it a model that has now sat unchanged for almost seven months while a parade of newer Gemini releases (the 3.5, 3.6 and 3.7 Flash tiers, and September 2's Gemini 3.8 Flash) shipped, scored, and passed it. Comparing these two is less a benchmark exercise than a bet on two very different release philosophies, and the freshness gap is the thing every spec sheet leaves out.

Seven months of preview, nine days of flagship

Gemini 3.1 Pro was Google's frontier reasoning model in February — "frontier" at the time, with strong software-engineering and agentic claims and a native multimodal foundation that still sets it apart. Google has not GA'd it since. In the seven months it has sat in preview, Anthropic has shipped two flagship generations: Claude Opus 5 on July 24, and now Claude Fable 5.1, which Anthropic positions as a Mythos-class tier above the Opus class and which took the top of the Artificial Analysis Intelligence Index at 66. When you buy into Gemini 3.1 Pro you are buying a model whose successor decisions Google has already effectively made elsewhere in its lineup; when you buy into Claude Fable 5.1 you are buying the model Anthropic is currently improving.

That asymmetry shows up in the small print. A frozen preview carries the risk that the capability you standardize on never gets the GA polish, the pricing rework, or the follow-on point releases. A nine-day-old flagship carries the opposite risk — that a 5.2 or a point release lands next month and moves the goalposts. Neither risk is avoidable; the point is that they point in opposite directions, and which one you can tolerate is a real input to this decision.

The spec sheet, side by side

• Price — Claude Fable 5.1 $10.00 / $50.00 per 1M vs Gemini 3.1 Pro $2.00 / $12.00 per 1M up to 200K input tokens, then $4.00 / $18.00 above 200K. Roughly 5x cheaper on input and 4x on output at the base tier.

• Context / max output — both 1M-token context; Claude Fable 5.1 outputs up to 128K vs Gemini 3.1 Pro's ~65K ceiling.

• Inputs — Claude Fable 5.1 takes text and image. Gemini 3.1 Pro takes text, image, audio, video, and PDF, with native Google Search grounding.

• Independent score — Claude Fable 5.1 at 66 on the Artificial Analysis Intelligence Index (max effort) vs Gemini 3.1 Pro in the upper 40s (it launched at 57 and has drifted down as AA re-ran its evals).

• Speed — Gemini 3.1 Pro streams around 540 tokens/sec per OrcaRouter's own telemetry, against the 67.2 tokens/sec Artificial Analysis measured for Claude Fable 5.1 — a large throughput gap in the other direction.

• Status — Claude Fable 5.1 stable, GA, released 2026-09-01; Gemini 3.1 Pro Preview, released 2026-02-19, not GA'd as of mid-September 2026.

The benchmark picture, labeled honestly

On raw capability the two models are not in the same tier, and the independent index says so more cleanly than any vendor slide. Claude Fable 5.1's 66 on the Artificial Analysis Intelligence Index (max effort) is the highest score AA has recorded — though AA's model page for the default-fallback configuration Anthropic serves lists 53, still #1 of 201 — while Gemini 3.1 Pro sits in the upper 40s. On Anthropic-reported component benchmarks, Fable 5.1 posts 55.8% on Terminal-Bench 4.0 and 52.6% on Terminal-Bench-Science 0.1 — scores no one has yet attributed to Gemini 3.1 Pro in any independent test. On the widely-corroborated SWE-bench-style rows, Gemini 3.1 Pro's February-era claims (about 75% on SWE-bench Verified per third-party trackers) trail the current Claude generation by a wide margin.

The honest caveat runs the other way on everything multimodal. Gemini 3.1 Pro reads native audio, native video, and PDFs, and it has Google Search built into the model — capabilities Claude Fable 5.1 does not offer at all, since Anthropic's flagship takes text and images only. On long-document, long-video, or search-grounded work, Gemini 3.1 Pro is not merely cheaper; it is the only one of the two that can do the task in a single call. That is a real edge, and it is easy to underweight when every benchmark you look at is text-and-code.

The price cliff hides the real cost story

Gemini 3.1 Pro's headline $2 / $12 rate only applies below 200K tokens of input per request. Above that, the whole request reprices at $4.00 in and $18.00 out — a doubling on input that makes long-context Gemini calls far less cheap than the marketing number implies. Claude Fable 5.1's $10 / $50 is flat at every length, with cache reads at $0.25 that get cheaper relative to its own list price the more a session re-reads context. The crossover is workload-shaped: short, high-volume multimodal calls favor Gemini 3.1 Pro by a mile; long agent sessions that repeatedly re-read a big context favor Claude Fable 5.1's cache economics even before capability enters the argument.

Who should pick which

Pick Gemini 3.1 Pro when the workload is genuinely multimodal or search-grounded, when throughput at low price matters more than the top of the reasoning curve, and when you are comfortable standardizing on a model Google has left in preview for seven months — ideally with a migration path to whatever Google GA's next. Pick Claude Fable 5.1 when you need the current reasoning ceiling, when you are building long-horizon agents where the 128K output ceiling and the $0.25 cache reads matter, and when you would rather bet on a model its vendor is actively improving. Both are on OrcaRouter at their providers' list prices with zero markup, so the routing-DSL version of this decision is cheap to change later: send multimodal and high-volume traffic to Gemini 3.1 Pro, escalate the hard reasoning and agentic traffic to Claude Fable 5.1, and re-tune the split the day Google finally GA's its Pro tier — or the day Anthropic ships whatever comes after Fable 5.1.