
Ideogram 4.6 vs Ideogram 4.5: What the Upgrade Actually Buys
- OrcaNEWOrcaCyber: Hy4 Flash Abliterated / Uncensored2026-11-18$0.07 / $0.28 per 1M tokens · 187 tok/s
- tencentNEWTencent: Hy4 Flash2026-11-16$0.02 / $0.07 per 1M tokens · 413 tok/s
- GoogleNEWGoogle: Gemini 4 Carbon2026-11-10$2.00 / $10.00 per 1M tokens · 87 tok/s
- AnthropicNEWAnthropic: Claude Fable 5.22026-11-09$8.00 / $40.00 per 1M tokens · 119 tok/s
- orcaNEWOrca: OrcaC2 1.02026-11-079 tok/s
- OrcaOrcaCyber: GLM 5.3 Abliterated / Uncensored2026-11-04$3.00 / $5.00 per 1M tokens · 89 tok/s
- OrcaOrcaCyber: Hy4 Abliterated / Uncensored2026-11-04$0.80 / $1.60 per 1M tokens · 135 tok/s
- OrcaOrcaCyber: Kimi K3.1 Pro Abliterated / Uncensored2026-11-04$2.00 / $5.00 per 1M tokens · 78 tok/s
- tencentTencent: Hy42026-11-04$0.83 / $2.50 per 1M tokens · 116 tok/s
- OrcaOrcaCyber: MiniMax M3.1 Flash Abliterated / Uncensored2026-10-29$0.80 / $1.60 per 1M tokens · 114 tok/s
- OrcaOrcaCyber: GLM 5.3 Flash Abliterated / Uncensored2026-10-27$0.80 / $1.60 per 1M tokens · 66 tok/s
- OrcaOrcaCyber: Qwen3.8 Flash Next Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 91 tok/s
- OrcaOrcaCyber: MiMo V2.6 Flash Abliterated / Uncensored2026-10-26$0.80 / $1.60 per 1M tokens · 41 tok/s
- MoonshotMoonshotAI: Kimi K3.1 Pro2026-10-2447Intelligence
- OrcaOrcaCyber: DeepSeek V4.1 Flash Abliterated / Uncensored2026-10-23$0.80 / $1.60 per 1M tokens · 212 tok/s
- minimaxMiniMax: M3.1 Flash2026-10-1929Intelligence
- OrcaOrca: OrcaCyber Ultra 1.02026-10-17$5.00 / $10.00 per 1M tokens · 111 tok/s
- deepseekDeepSeek: DeepSeek V4.1 Flash 10172026-10-1733Intelligence
- DeepSeekDeepSeek: DeepSeek V4.1 Flash 09102026-10-17$0.15 / $0.60 per 1M tokens · 168 tok/s
- DeepSeekDeepSeek: DeepSeek V4.1 Flash2026-10-1740Intelligence
The useful thing to know about Ideogram 4.6 before anything else is that, below 4K, it costs exactly what Ideogram 4.5 costs. Same four quality modes, same two endpoints, same per-image prices at 2K on every rung of the ladder. The upgrade is free unless you cross into 4,096 pixels. That single fact reframes the whole comparison, because it means the decision is not "is 4.6 worth the extra money" but "does anything in my workflow need the extra pixels, and is the model better at the things I already do."
Ideogram 4.5 shipped on 30 September 2026 as a precise-editing model with a fixed price. Ideogram 4.6 shipped on 19 November 2026 and broke that pricing model in two: it kept the 2K prices and added a 4K tier on top. Everything else — the endpoint names, the quality modes, the five-reference limit — carried over unchanged.
Two models, one price sheet below 4K

Put the two rate cards side by side and the only rows that differ are the 4K ones. Ideogram 4.5 charges one price per image and says so explicitly: the price "does not change with prompt length, the number of reference images, or the output resolution." Ideogram 4.6 charges by quality and by resolution, with everything at 2K or below billed at the 2K price.
• Very low — 4.5 $0.008; 4.6 $0.008 at 2K, no 4K option at either model's floor
• Low — 4.5 $0.03 flat; 4.6 $0.03 at 2K, $0.06 at 4K
• Medium — 4.5 $0.06 flat; 4.6 $0.06 at 2K, $0.12 for 4K generation and $0.20 for a 4K edit
• High — 4.5 $0.10 prompt-only and $0.22 with references or Precise Edit; 4.6 $0.10 generate and $0.22 edit at 2K, $0.20 generate and $0.60 edit at 4K
Read down the 2K column and the two models are identical at every rung. The divergence is entirely above 2K, and it is steep: a High-quality 4K edit on 4.6 is $0.60, which is 2.7 times the same edit at 2K and six times a 2K generation at the same quality setting. Medium is worse in relative terms — a 4K edit at Medium costs more than a 2K edit at High.
That asymmetry is the actual news in this release for anyone already paying Ideogram. You are not being asked to pay more for the model. You are being asked to pay more for pixels, and the multiplier only bites if your deliverable needs them.
What the 26 points are worth

The benchmark gap between the two models is real but smaller than the headline suggests, and the shape of it matters more than the size. Rapidata's Multi-step Image Edit board places 4.6 first at 1,892 and 4.5 second at 1,866 — 26 points apart on a Bradley–Terry fit rescaled to Elo, from 413,667 human responses across 76,714 pairwise matches on 199 photographs of real people.
The intervals are the part to hold onto. 4.6 is reported with a 32-point interval and 4.5 with a 30-point interval, so 4.6's lower bound of roughly 1,860 sits below 4.5's upper bound of roughly 1,896. The point estimates rank 4.6 first and the board is right to print that ranking. But at 95% confidence the two models are not cleanly separated on this board, and no amount of quoting the 26-point gap changes that.
There is a second reason not to over-read it. The two rows are not the same configuration: 4.6 is evaluated as Precise Edit at High Quality and 4.5 as High Quality with High Precision. Both are the top setting each model offers, but "High Precision" is a 4.5-era label that does not correspond to any mode in the 4.6 API. A family upgrade measured across two different top configurations is a fair ranking of the two products as shipped. It is not a measurement of what changed between them.
The five tasks where 4.5 gives ground
The vendor's own comparison chart is more useful here than the leaderboard, because it breaks the improvement into the tasks a workflow is actually made of, and because the gaps are far too large to be noise. These are vendor-reported rates for how often each model delivered a usable result on the same requests at High quality.
• Character consistency — 85% on 4.6, up from 69% on 4.5. A sixteen-point gain, and the one that matters most for portrait or talent work where the face has to survive a sequence.
• Exact color — 83%, up from 46%. The largest gap by a wide margin, and the one with the clearest commercial reading: brand-colour work either hits the specified value or fails.
• Product detail — 86%, up from 61%. Repairing a small defect in a product photograph.
• Product shots — 88%, up from 66%. Staging, restyling and relabelling a product.
• Masked edits — 90%, up from 82%. The smallest gain of the five, on the task where a mask already constrains the model.
Sort those by improvement and a pattern emerges that the leaderboard hides. The gains are largest exactly where the model has to hold a specific value it was given — a hex, a product's details, a person's identity — and smallest where the request itself narrows the answer space. That is the signature of a model that got better at obedience rather than one that got better at imagination, and it predicts which workflows will feel the difference on day one and which will not.
The usual caveat applies and should not be skipped: "usable result" has no published definition, no annotator count, and no statement of whether judgments were blind. The direction is credible because it lines up with the Rapidata result. The magnitudes are the vendor's.
Reading 4.5's independent scores

Ideogram 4.5 has something 4.6 does not yet have: a full independent record. On Artificial Analysis it holds rank 37 of the text-to-image board at 1,017 Elo from 3,549 samples, priced at $100 per thousand images, and rank 34 on the editing board at 1,065 Elo from 4,003 samples at $220 per thousand. Those are modest positions on a crowded board — the top of the editing board sits above 1,150 — and they describe a competent mid-tier model rather than a leader.
The listed $220 per thousand for editing is worth pausing on, because it is close to double the $0.22 the vendor charges for a High-quality Precise Edit. The board's figure reflects whichever access path it measured, and access-path pricing is exactly the kind of thing that varies between a direct vendor call and a resold one. Taken at face value it says the same thing the vendor sheet says: High-quality edits are where the cost concentrates, and they concentrate there on both models.
What the independent record cannot yet tell you is whether 4.6 improves on 1,065. There is no 4.6 row on either Artificial Analysis board. Until there is, the only third-party measurement of the new model is the Rapidata board, and the comparison against 4.5 rests on that one source.
Migrating without a rewrite
The mechanical half of this upgrade is close to trivial, which is not something that can be said of most model migrations. The endpoint structure is unchanged: /v2/image/generate/ideogram-4-5 becomes /v2/image/generate/ideogram-4-6, and /v2/image/precise-edit/ideogram-4-5 becomes /v2/image/precise-edit/ideogram-4-6. The four quality modes keep their names. The five-reference limit on Generate is the same. The optional mask on Precise Edit is the same. Authentication is the same header.
The one behavioural difference to plan for is that 4.6's Precise Edit preserves the source image's dimensions, and 4.6's Very low mode is capped at 2K even though the model supports 4K. If your integration pins a resolution per request, that pin is now a price decision rather than a formatting one.
On routing: neither Ideogram model is on OrcaRouter's model list today, so there is no shortcut here and no reason to expect our per-call price to differ from Ideogram's own. What a routing layer is genuinely good for in a migration like this is the other half of the pipeline. If you pair Ideogram edits with a generation model we do host — Nano Banana 2, Nano Banana Pro or the gpt-image-2.5 pair — that half can sit behind automatic failover, so a provider incident does not take the whole job down while you are still validating the new editor.
When to stay on 4.5
Staying is the right call in two situations, and both are common enough to name.
The first is a workflow that never exceeds 2K and edits rarely. You would be paying zero extra for 4.6 — the prices match — but you would also be gaining very little, because the drift improvement compounds across turns and a single-edit workflow never gets to the third turn. The vendor's own model guide says as much: it recommends comparing 4.6 against 4.5 before migrating an established integration, and positions 4.6 specifically for "precise editing workflows and output up to 4K."
The second is a workflow with a hard cost ceiling per asset. A High-quality 4K edit at $0.60 changes the unit economics of anything produced at volume, and the 4K tier is the only thing 4.6 sells that 4.5 does not. If 4K is not in the brief, the tier is not in the budget either.
For everyone else — long edit chains, brand-colour work, product photography, anything where the same source image gets refined across many turns — the upgrade costs nothing at the resolution you are already using, and the evidence, thin as it is around the edges, points one direction. Move, and test on your own worst chain rather than the gallery.
